Cable core material identification method based on neural network

Abstract

The invention discloses a cable core material identification method based on a neural network. The cable core material identification method comprises the following steps: step 1, acquiring an image of a cable core material; step 2, preprocessing the image; step 3, obtaining an edge contour of the cable core material by using an edge detection algorithm; step 4, extracting color features of the cable core material; step 5, training a cable core material identification model based on a neural network architecture; and step 6, processing the to-be-identified image of the cable core material by using the methods from the step 2 to the step 5, identifying and classifying the to-be-identified image by using the cable core material identification model, and outputting a classification result. The method is simple in structure and reasonable in design, and in the edge detection algorithm, the maximum between-class variance is used for searching the optimal threshold value; when the clustering center is updated, a momentum item is introduced; the cable core material identification model is constructed based on an ART model, can maintain stable identification performance under different illumination conditions, image angles and background interferences, and has good robustness.

Description

Technical Field

[0001] The invention belongs to the technical field of image recognition, and in particular relates to a cable core material recognition method based on a neural network. Background Art

[0002] Copper core cable usually refers to a cable made of higher purity copper material, such as red copper or electrolytic copper. It has excellent conductivity, low resistivity, high transmission efficiency, good corrosion resistance, high tensile strength and bending strength, and is not easy to break. It is suitable for power transmission, communication, electronic equipment and other fields, especially those requiring high-efficiency transmission and high reliability.

[0003] In the power cable market, some unscrupulous merchants may use other materials to pass off as copper core cables in order to reduce costs. Copper core cables are often passed off as copper core cables by the following materials: 1. Brass or steel: Brass is an alloy of copper and zinc, and its color is affected by the zinc content. The conductivity of these materials is not as good as copper, but the price is lower. Merchants may wrap a thin layer of red copper on the outer layer to deceive consumers. 2. Recycled copper or impure copper: Recycled copper is a product obtained by re-smelting recycled scrap copper. The content of recycled copper is unstable, the resistivity is large, and the conductivity is not as good as high-quality copper. However, due to the relatively low price, some unscrupulous merchants may use these materials to make cables.

[0004] Recycled copper, brass and high-quality copper all contain copper content, and these differences may not be obvious during certain physical tests, especially during initial quick tests such as a simple conductivity test or visual inspection.

[0005] However, the color of recycled copper may be duller and contain some impurities; the color of brass varies from golden yellow to reddish yellow, affected by the zinc content; and the copper material with higher purity appears purple-red or red-orange with a strong metallic luster. These differences make them have different visual effects in practical applications. Cable core materials can be distinguished based on color characteristics. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a cable core material identification method based on a neural network in view of the deficiencies in the above-mentioned prior art. The method has a simple structure and a reasonable design. In the edge detection algorithm, the maximum inter-class variance is used to find the optimal threshold; when updating the cluster center, a momentum term is introduced; the cable core material identification model is constructed based on the ART model, which can maintain stable recognition performance under different lighting conditions, image angles and background interference, and has good robustness.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: a cable core material identification method based on neural network, characterized in that it includes the following steps:

[0008] Step 1: Collect images of cable core materials;

[0009] Step 2: preprocess the image;

[0010] Step 3: Use edge detection algorithm to obtain the edge contour of the cable core material;

[0011] Step 4: Extract the color characteristics of the cable core material:

[0012] Step 401, converting the image within the edge contour from the RGB color space to the HSV color space;

[0013] Step 402: Use the K-means clustering algorithm to perform color clustering on the image within the edge contour:

[0014] Step 403, calculating the color features of k cluster centers;

[0015] Step 404, calculating the statistics of the color features of k cluster centers;

[0016] Step 5: Training a cable core material recognition model based on a neural network architecture;

[0017] Step 6: Use the methods of steps 2 to 5 to process the image of the cable core material to be identified, use the cable core material identification model to identify and classify the image to be identified, and output the material classification result of the cable core material.

[0018] The above-mentioned method for identifying cable core materials based on a neural network is characterized in that: in step 3, the specific method of using an edge detection algorithm to obtain the edge contour of the cable core material includes:

[0019] Step 301, converting the image into a grayscale image, calculating the gradient of the grayscale image, calculating the gradient value and the gradient direction angle according to the gradient; merging the gradient direction angle into n directions;

[0020] Step 302, traverse the pixel points in n directions respectively, and determine whether the current pixel point is the maximum value of the surrounding pixel points with the same directional gradient; if so, retain the point, otherwise return its pixel value to zero;

[0021] Step 303: Calculate the inter-class variance g(t), g(t)=w0·w1·(u0-u1)2, and find the threshold TH corresponding to max[g(t)], where w0 represents the proportion of pixels whose grayscale value is less than the threshold TH, w1 represents the proportion of pixels whose grayscale value is greater than or equal to the threshold TH, u0 represents the foreground grayscale mean, and u1 represents the background grayscale mean;

[0022] Step 304: Set the threshold TH as the high threshold and b·TH as the low threshold, 0>b>1;

[0023] Step 305: Mark edge pixels according to the relationship between the pixel gradient value and the low threshold and the high threshold to obtain an edge contour.

[0024] The above-mentioned cable core material identification method based on neural network is characterized in that: in step 402, the specific method of using the K-means clustering algorithm to perform color clustering on the image within the edge contour includes:

[0025] Step 4021: randomly select HSV values ​​of K pixels as initial cluster centers;

[0026] Step 4022: Calculate pixel x i The distance to K cluster centers is calculated and assigned to the cluster center C that is closest to it. k ;

[0027] Step 4023, according to the formula Calculate the position μ of the kth cluster center at the t+1th iteration k，t+1 , C k represents the set of all pixels of the kth cluster center, |C k Represents the set C k The number of k，t represents the momentum term, β represents the weight of the momentum term, and α represents the adjustment weight;

[0028] Step 4024: Repeat steps 4022 and 4023 until a preset number of iterations is reached.

[0029] The above-mentioned cable core material identification method based on neural network is characterized by: K=3.

[0030] The above-mentioned cable core material identification method based on neural network is characterized in that: in step 403, the specific method of calculating the color characteristics of k cluster centers includes:

[0031] Color feature extraction: According to the formula Calculate the color features of k cluster centers separately in represents the H-mean of the k-th cluster center, h i represents the H value of the i-th pixel in the k-th cluster center; where S k represents the S-mean of the k-th cluster center, s i represents the S value of the i-th pixel in the k-th cluster center; where V k represents the V mean of the kth cluster center, v i Represents the V value of the i-th pixel in the k-th cluster center.

[0032] The above-mentioned cable core material identification method based on neural network is characterized in that: the image of the cable core material collected in step one includes images of cable core materials of different lighting scenes, different brands, and different materials; different lighting scenes refer to micro-exposure scenes, balanced light scenes, shadow light scenes, weak light scenes and no light scenes.

[0033] The above-mentioned cable core material identification method based on neural network is characterized in that: in step 2, the image is preprocessed, including converting the image into numerical data and performing illumination balancing processing on the image.

[0034] The above-mentioned cable core material identification method based on neural network is characterized by: a cable core material identification model based on a neural network architecture, specifically a cable core material identification model based on ART.

[0035] The above-mentioned cable core material identification method based on neural network is characterized in that: the cable core material identification model based on ART includes an input layer, a feature extraction layer, an ART layer and an output layer, wherein the output layer includes at least 3 neurons.

[0036] The above-mentioned cable core material identification method based on neural network is characterized in that the activation function of the cable core material identification model based on ART uses the Tanh function.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] 1. The invention has a simple structure, reasonable design, and is easy to implement and operate.

[0039] 2. In the edge detection algorithm of the present invention, the maximum inter-class variance is used to find the optimal threshold value, which can divide the image into two parts, the foreground and the background, so that the inter-class variance of the two parts is maximized, avoiding the subjectivity and inaccuracy of manually setting the threshold value, and having strong adaptability to different lighting conditions, noise levels and image contents, thereby accurately detecting the edge.

[0040] 3. When updating the cluster center, the present invention introduces a momentum term. The cluster center of the new iteration depends not only on the distribution of the current data points, but also on the position of the cluster center in the previous iteration. The direction and size of the previous iteration update are used to correct the current update direction, making the iteration process smoother and more stable.

[0041] 4. The cable core material recognition model of the present invention is constructed based on the ART model. Through its unique adaptive resonance mechanism, it can dynamically adjust network parameters when facing a variety of cable core material images to achieve high-precision material classification; it can maintain stable recognition performance under different lighting conditions, image angles and background interference, and has good robustness.

[0042] In summary, the present invention has a simple structure and a reasonable design. In the edge detection algorithm, the maximum inter-class variance is used to find the optimal threshold; when updating the cluster center, the momentum term is introduced; the cable core material recognition model is constructed based on the ART model, which can maintain stable recognition performance under different lighting conditions, image angles and background interference, and has good robustness.

[0043] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The figure is a flow chart of the method of the present invention.

[0045] Figure 2 It is a structural schematic diagram of the present invention.

[0046] Description of reference numerals:

[0047] Computing module—1; Camera—2. DETAILED DESCRIPTION

[0048] The method of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments of the present invention.

[0049] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0050] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0051] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0052] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0053] like Figure 1 As shown, a cable core material identification method based on a neural network of the present invention comprises the following steps:

[0054] Step 1: Collect images of cable core materials;

[0055] It should be noted that the images of the cable core material are collected using camera 2 or a high-definition camera, and the images of the cable core material include images of cable core materials of different lighting scenes, different brands, and different materials. Different lighting scenes refer to micro-exposure scenes, balanced light scenes, shadow light scenes, weak light scenes, and no light scenes.

[0056] Step 2: preprocess the image;

[0057] The image is processed using the computing module 1.

[0058] First, the image is converted into numerical data and the image is processed for illumination balance. Since the image is taken under different lighting scenes, the illumination balance process can make the image maintain consistent visual effects and recognition performance under different lighting environments, and adapt to the complex lighting conditions in outdoor detection and recognition scenes.

[0059] Step 3: Use edge detection algorithm to obtain the edge contour of the cable core material;

[0060] Step 301: convert the image into a grayscale image, calculate the gradient of the grayscale image, calculate the gradient value and the gradient direction angle according to the gradient; and merge the gradient direction angle into gradient direction angle directions.

[0061] In actual use, n=6, and the 6 gradient direction angles are:

[0062] First gradient direction angle: 0 degrees;

[0063] Second gradient direction angle: 60 degrees;

[0064] The third gradient direction angle: 120 degrees;

[0065] Fourth gradient direction angle: 180 degrees;

[0066] Fifth gradient direction angle: 240 degrees;

[0067] Sixth gradient direction angle: 300 degrees.

[0068] Step 302, traverse the pixel points in n directions respectively, and determine whether the current pixel point is the maximum value of the surrounding pixel points with the same directional gradient; if so, retain the point, otherwise return its pixel value to zero;

[0069] Step 303, calculate the inter-class variance g(t), g(t) = w0·w1·(u0-u1)2, and find the threshold TH corresponding to max[g(t)], where w0 represents the proportion of pixels whose grayscale values ​​are less than the threshold TH, w1 represents the proportion of pixels whose grayscale values ​​are greater than or equal to the threshold TH, u0 represents the foreground grayscale mean, and u1 represents the background grayscale mean.

[0070] The threshold TH corresponding to the maximum inter-class variance is the optimal threshold, which can segment the image into two parts, the foreground and the background, so that the inter-class variance of the two parts is maximized, avoiding the subjectivity and inaccuracy of manually setting the threshold.

[0071] The maximum inter-class variance is used in the edge detection algorithm, which has strong adaptability to different lighting conditions, noise levels and image contents, thereby accurately detecting edges.

[0072] It should be noted that w0 is referred to as the background pixel ratio, and w1 is referred to as the foreground pixel ratio. If the threshold TH is set differently, the divisions of w0 and w1 will be different, and the calculation results of u0 and u1 will also be different.

[0073] Step 304: Set the threshold TH as the high threshold, and b·TH as the low threshold, 0>b>1. In a possible embodiment, b=0.3 or 0.5.

[0074] Step 305: Mark edge pixels according to the relationship between the pixel gradient value and the low threshold and the high threshold to obtain an edge contour;

[0075] If the pixel gradient value is greater than a high threshold, it is marked as a strong edge pixel.

[0076] If the gradient value of a pixel is less than the high threshold but greater than the low threshold, it is marked as a weak edge pixel.

[0077] If the gradient value of a pixel is less than a low threshold, it is suppressed as a non-edge pixel.

[0078] Step 4: Extract the color characteristics of the cable core material:

[0079] Step 401, color space conversion: convert the image within the edge contour from the RGB color space to the HSV color space.

[0080] In the RGB color space, color is determined by three components: red (R), green (G), and blue (B). These three components are closely related to brightness. As long as the brightness changes, the three components will also change, making it relatively complicated to extract color features in the RGB color space.

[0081] In the HSV color space, since brightness and color information are separated, it is easier to extract color features. In the HSV model, color is decomposed into three components: hue, saturation, and value.

[0082] Step 402: Color cluster analysis:

[0083] Step 4021: randomly select HSV values ​​of K pixels as initial cluster centers;

[0084] Step 4022: Calculate pixel x i The distance to K cluster centers is calculated and assigned to the cluster center C that is closest to it. k ;

[0085] Step 4023, according to the formula Calculate the position μ of the kth cluster center at the t+1th iteration k，t+1 , Ck represents the set of all pixels of the kth cluster center, |C k Represents the set C k The number of k，t represents the momentum term, β represents the weight of the momentum term, and α represents the adjustment weight;

[0086] When updating the cluster center, the momentum term is introduced to use the direction and size of the previous iteration update to correct the current update direction, making the iteration process smoother and more stable. The cluster center of the new iteration depends not only on the distribution of the current data points, but also on the location of the cluster center in the previous iteration. k，t The size of can be adjusted by setting the weight α of the momentum term.

[0087] Step 4024: Repeat steps 4022 and 4023 until a preset number of iterations is reached.

[0088] In one possible embodiment, when color clustering is performed on the color of the cable core material, the core material color clearly includes three categories: recycled copper, brass, and high-quality copper, that is, theoretically, the number of clusters should be set to 3. This is because the basic idea of ​​the K-means clustering algorithm is to cluster the samples in the data set into K clusters, each cluster is represented by a cluster center, and each sample point belongs to the cluster where the cluster center closest to it is located.

[0089] Step 403, color feature extraction: According to the formula Calculate the color features of k cluster centers separately in represents the H-mean of the k-th cluster center, h i represents the H value of the i-th pixel in the k-th cluster center; represents the S-mean of the k-th cluster center, s i represents the S value of the i-th pixel in the k-th cluster center; represents the V mean of the kth cluster center, v i Represents the V value of the i-th pixel in the k-th cluster center.

[0090] In a possible embodiment, the H average value of recycled copper is 10-30, the H average value of brass is 30-60, and the H average value of high-quality copper is 300-330.

[0091] It should be noted that the saturation of the core material is generally 1. Moreover, the surface of the core material that has been polished and ground usually has a higher lightness value, making it look brighter and more shining. The surface of the untreated core material may be relatively dull and have a lower lightness value. Therefore, in a possible embodiment, when extracting color features, only the H mean is calculated to reduce the amount of calculation.

[0092] Step 404: Calculate the statistics of the color features of the k cluster centers. In a possible embodiment, the statistics of the color feature refers to the mode.

[0093] Step 5: Training a cable core material recognition model based on a neural network architecture;

[0094] A cable core material recognition model based on a neural network architecture, specifically a cable core material recognition model based on ART.

[0095] Although the H mean of HSV can be used as a preliminary basis for the color classification of power cables, inputting the H mean, S mean and V mean into the cable core material recognition model for classification and identification can further improve the accuracy and reliability of classification. Even when encountering new and unseen cable color samples, the neural network can classify and identify them based on the learned knowledge, thereby achieving more accurate classification and identification of power cables.

[0096] Step 6. Model usage: Use the methods of steps 2 to 5 to process the image of the cable core material to be identified, use the cable core material identification model to identify the image to be identified, and output the material classification result of the cable core material.

[0097] The cable core material recognition model based on ART includes an input layer, a feature extraction layer, an ART layer and an output layer, wherein the output layer includes at least 3 neurons. The feature extraction layer usually contains multiple convolutional layers and pooling layers. The ART layer dynamically learns and updates its internal representation by adaptively adjusting its parameters such as weights and biases to achieve accurate classification of the cable core material. The activation function uses the Tanh function, and the output of the Tanh function is centered on 0, which is conducive to centralizing the data and making the gradient propagation more stable. The output layer is responsible for outputting the final cable core material classification results. In a possible embodiment, the classification results include three material classifications: recycled copper, brass and high-quality copper, and contain 3 neurons, each neuron corresponding to a possible material category.

[0098] In one possible embodiment, the cable core material recognition model is constructed based on the ART adaptive resonance theory model, which is particularly suitable for processing complex and variable image recognition tasks, such as the material classification of cable core materials. The cable core material recognition model uses the powerful pattern recognition and memory capabilities of the ART network to efficiently extract and match the input image features. Through its unique adaptive resonance mechanism, the cable core material recognition model can dynamically adjust network parameters when facing a variety of cable core material images to achieve high-precision material classification.

[0099] During the recognition process, the cable core material recognition model compares the features of the input image with the known cable core material features stored in the memory unit. Once the best match is found, the model will output the corresponding cable core material classification results, including three material classifications: recycled copper, brass, and high-quality copper.

[0100] It is worth noting that the cable core material recognition model based on the ART model can maintain stable recognition performance under different lighting conditions, image angles and background interference, and has good robustness.

[0101] The above description is only an embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A cable core material identification method based on neural network, characterized in that: The following steps are involved: Step 1: Collect images of cable core materials; Step 2: preprocess the image; Step 3: Use edge detection algorithm to obtain the edge contour of the cable core material; Step 4: Extract the color characteristics of the cable core material: Step 401, converting the image within the edge contour from the RGB color space to the HSV color space; Step 402: Use the K-means clustering algorithm to perform color clustering on the image within the edge contour: Step 403, calculating the color features of k cluster centers; Step 404, calculating the statistics of the color features of k cluster centers; Step 5: Training a cable core material recognition model based on a neural network architecture; Step 6: Use the methods of steps 2 to 5 to process the image of the cable core material to be identified, use the cable core material identification model to identify and classify the image to be identified, and output the material classification result of the cable core material.

2. A method for identifying cable core material based on a neural network according to claim 1, characterized in that: In step 3, the specific method of using the edge detection algorithm to obtain the edge contour of the cable core material includes: Step 301, converting the image into a grayscale image, calculating the gradient of the grayscale image, calculating the gradient value and the gradient direction angle according to the gradient; merging the gradient direction angle into n directions; Step 302, traverse the pixel points in n directions respectively, and determine whether the current pixel point is the maximum value of the surrounding pixel points with the same directional gradient; if so, retain the point, otherwise return its pixel value to zero; Step 303: Calculate the inter-class variance g(t), g(t)=w0·w1·(u0-u1)2, and find the threshold TH corresponding to max[g(t)], where w0 represents the proportion of pixels whose grayscale value is less than the threshold TH, w1 represents the proportion of pixels whose grayscale value is greater than or equal to the threshold TH, u0 represents the foreground grayscale mean, and u1 represents the background grayscale mean; Step 304: Set the threshold TH as the high threshold and b·TH as the low threshold, 0>b>1; Step 305: Mark edge pixels according to the relationship between the pixel gradient value and the low threshold and the high threshold to obtain an edge contour.

3. A cable core material identification method based on a neural network according to claim 1, characterized in that: In step 402, the specific method of using the K-means clustering algorithm to perform color clustering on the image within the edge contour includes: Step 4021: randomly select HSV values ​​of K pixels as initial cluster centers; Step 4022: Calculate pixel x i The distance to the K cluster centers is calculated and assigned to the cluster center C that is closest to it. k ; Step 4023, according to the formula Calculate the position μ of the kth cluster center at the t+1th iteration k，t+1 , C k represents the set of all pixels of the kth cluster center, |C k Represents the set C k The number of k，t represents the momentum term, β represents the weight of the momentum term, and α represents the adjustment weight; Step 4024: Repeat steps 4022 and 4023 until a preset number of iterations is reached.

4. A cable core material identification method based on a neural network according to claim 3, characterized in that: K＝3。 5. A method for identifying cable core material based on neural network according to claim 1, characterized in that: In step 403, the specific method of calculating the color features of k cluster centers includes: Color feature extraction: According to the formula Calculate the color features of k cluster centers separately in represents the H-mean of the k-th cluster center, h i represents the H value of the i-th pixel in the k-th cluster center; represents the S-mean of the k-th cluster center, s i represents the S value of the i-th pixel in the k-th cluster center; represents the V mean of the kth cluster center, v i Represents the V value of the i-th pixel in the k-th cluster center.

6. A method for identifying cable core material based on a neural network according to claim 1, characterized in that: The images of the cable core material collected in step one include images of cable core materials of different lighting scenes, different brands, and different materials; different lighting scenes refer to micro-exposure scenes, balanced light scenes, shadow light scenes, weak light scenes, and no light scenes.

7. A cable core material identification method based on a neural network according to claim 1, characterized in that: In step 2, the image is preprocessed, including converting the image into numerical data and performing illumination balancing on the image.

8. A method for identifying cable core materials based on a neural network according to claim 1, characterized in that: A cable core material recognition model based on a neural network architecture, specifically a cable core material recognition model based on ART.

9. A method for identifying cable core material based on a neural network according to claim 8, characterized in that: The cable core material recognition model based on ART includes an input layer, a feature extraction layer, an ART layer and an output layer, wherein the output layer includes at least 3 neurons.

10. A method for identifying cable core material based on a neural network according to claim 8, characterized in that: The activation function of the cable core material recognition model based on ART uses the Tanh function.

Images