Method for identifying fin end part of heat exchanger in refrigeration field

Abstract

The invention discloses a method for identifying a fin end part of a heat exchanger in the refrigeration field, which comprises the following steps of: acquiring a detection image set of an air conditioner fin end part image, and acquiring a fin end plate hole Mark point pixel position coordinate, a fin end plate hole pixel position coordinate and a physical coordinate by using a template matching tool; marking the position of the guide pin in the fin end plate hole in the detection image set to obtain a marked image set, and inputting the marked image set into a YOLO neural network for training to obtain an end plate hole guide pin recognition model; inputting a detection image set into the model to obtain a guide needle coordinate; comparing the coordinates of the guide pin with the coordinates of the pixel position of the fin end plate hole, calculating to obtain an offset value of the guide pin, and guiding the physical coordinates of the pin according to the offset value and the physical coordinates of the fin end plate hole; according to the method, all the guide pins and hole coordinates on the fins can be output at a time, the speed is high, the calculation result precision is higher, the recognition effect is more stable, the accurate guide pin physical coordinates can be obtained, and the accuracy of automatically inserting the heat exchange tubes is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fin end plate hole identification, and particularly to a method for identifying the ends of fins of a heat exchanger in the refrigeration field. Background Art

[0002] During the processing of air conditioners, it is necessary to process the fins into which the heat exchange tubes are inserted and stack multiple fins. The fins are often provided with multiple hole positions, and guiding pins are arranged in the hole positions. After the fins are processed, an end plate is placed at one end of the fins. The end plate is also provided with multiple hole positions that are the same as the positions of the fin holes. The heat exchange tubes are inserted into the hole positions of the fins through the end plate to complete the assembly of the heat exchanger;

[0003] However, when positioning the fin holes of the heat exchanger, there are the following technical difficulties: 1. Since there are usually 20 to 30 fin holes, it is cumbersome to identify and calculate each hole in turn, and the calculation benchmarks for each hole are not unified. The debugging period is long, the workload is large, and the overall data is not standardized and accurate enough, which is not convenient for later maintenance; 2. Since there are guiding pins in the fin holes during the assembly of the air conditioner fins, the depths of the guiding pins inserted into the fin holes are different, and the fins themselves have a certain amount of deformation during the assembly process. The traditional single Mark point identification method cannot meet the requirement of accurately positioning the actual positions of the guiding pins on each fin hole. Since the fins are relatively soft, identifying local Mark points cannot eliminate the overall deformation error of the fins; 3. When there are guiding pins in the identified fin holes, the end plate holes are blocked, and the image difference from when the holes are empty is large. The hole boundaries are not clear enough, the visual imaging effect is not good, and the accuracy is insufficient. Therefore, the existing fin hole positioning methods cannot obtain a unified calculation benchmark for the fin holes, cannot accurately position the positions of the guiding pins in each fin, have poor imaging, insufficient accuracy, and it is easy to make mistakes when automatically inserting the heat exchange tubes according to the positions of the fin holes and the guiding pins, which cannot meet the production requirements of the products. Summary of the Invention

[0004] The present invention provides a method for identifying the ends of fins of a heat exchanger in the refrigeration field to overcome the technical problems that the existing fin hole positioning methods cannot obtain a unified position benchmark for the fin holes, cannot accurately position the positions of the guiding pins in each fin, have poor imaging, and insufficient accuracy.

[0005] To achieve the above object, the technical solution of the present invention is:

[0006] A method for identifying the ends of fins of a heat exchanger in the refrigeration field, comprising:

[0007] S1: Obtain an image set of the end of the air conditioner fin, preprocess the image of the end of the air conditioner fin to obtain a detection image set, and the detection image includes fin end plate holes and guiding pins in the fin end plate holes;

[0008] S2: Use a template matching tool to obtain the pixel position coordinates of the Mark points in the fin end plate holes in the detection image set. Obtain the pixel position coordinates of the fin end plate holes based on the pixel position coordinates of the Mark points, and then obtain the physical coordinates of the fin end plate holes based on the pixel position coordinates of the fin end plate holes.

[0009] S3: Mark the position of the guide pins in the fin end plate holes in the detection image set to obtain a marked image set. Input the marked image set into the YOLO neural network for training to obtain a fin end plate hole guide pin recognition model.

[0010] S4: Input the detection image set into the fin end plate hole guide pin recognition model to obtain the guide pin coordinates in the fin end plate holes in the detection image.

[0011] S5: Compare and calculate the guide pin coordinates in the fin end plate holes with the pixel position coordinates of the corresponding fin end plate holes to obtain the offset value of the guide pin relative to the pixel position coordinates of the fin end plate holes. Obtain the physical coordinates of the guide pin in the fin end plate holes in the final actual situation based on the offset value and the physical coordinates of the corresponding fin end plate holes.

[0012] Further, using a template matching tool to obtain the pixel position coordinates of the Mark points in the fin end plate holes in the detection image set includes:

[0013] S21: Randomly select a detection image as a reference image. Use a template matching tool to select two end plate holes closest to the center and corners of the image on the reference image as two reference Mark points. The pixel position coordinates of the two reference Mark points are P1(x1, y1) and P2(x2, y2) respectively, where P1 is used as the origin.

[0014] S22: Randomly select another detection image, and re-obtain the pixel position coordinates of the two reference Mark points on this detection image, which are P′ N (x′ N , y′ N ) and P′ N+1 (x′ N+1 , y′ N+1 ), N ∈ 1, 2, 3…n, and P′ N is used as the origin.

[0015] S23: Use the affine transformation algorithm to calculate the pixel position coordinates of each fin end plate hole on the detection image based on the two sets of Mark point coordinates.

[0016] Further, using the affine transformation algorithm to calculate the pixel position coordinates of each fin end plate hole on the detection image based on the two sets of Mark point coordinates includes:

[0017] S231: Calculate the included angle between the pixel position coordinates of two reference Mark points in the detected image and the pixel position coordinates of two reference Mark points in the reference image, as shown in formula (1).

[0018]

[0019] Where tanθ represents the tan value of the included angle between the pixel position coordinates of two reference Mark points in the detected image and the coordinates of two reference Mark points in the reference image; θ is the included angle between the pixel position coordinates of two reference Mark points in the detected image and the pixel position coordinates of two reference Mark points in the reference image.

[0020] S232: Obtain the offset of the origin of the detected image relative to the origin of the reference image, as shown in formulas (2) and (3).

[0021] Δx N = x′ N - x1 (2)

[0022] Δy N = y′ N - y1 (3)

[0023] Where Δx N represents the offset of the abscissa, and Δy N represents the offset of the ordinate.

[0024] S233: Construct an affine transformation matrix for the fin end plate holes in the detected image based on the calculated included angle and offset, as shown in formula (4).

[0025]

[0026] S234: Obtain the pixel position coordinates of a certain fin end plate hole in the reference image after transformation in the detected image according to the affine transformation matrix, as shown in formula (5).

[0027]

[0028] Where O0(X0,Y0) represents the pixel position coordinates of the fin end plate hole in the reference image, and O N (X N ,Y N ) represents the pixel position coordinates of the fin end plate hole in the reference image after transformation in the detected image; X N and Y N are as shown in formulas (6) and (7).

[0029] X N = X0 cosθ - Y0 sinθ + Δx N (6)

[0030] Y N = X0 sinθ + Y0 cosθ + Δy N (7).

[0031] Further, obtaining the physical coordinates of the fin end plate holes according to the pixel position coordinates of the fin end plate holes includes:

[0032] S24. Calculating a conversion coefficient according to the physical distance between two adjacent fin end plate holes in the reference image and the pixel position coordinates of two reference Mark points, where the conversion coefficient is used to convert pixel position coordinates into physical coordinates, and the conversion coefficient is shown in formula (8),

[0033]

[0034] where K is the conversion coefficient and L is the physical distance between two adjacent fin end plate holes;

[0035] S25. Converting the pixel position coordinates O of the fin end plate holes in the detection image into physical coordinates according to the conversion coefficient and the origin P1(x1, y1) of the reference image N (X N , Y N ) as shown in formulas (9) and (10),

[0036] X UN = (X N - x1)·K (9)

[0037] Y UN = (Y N - y1)·K (10)

[0038] U N (X UN , Y UN ) represents the physical coordinates after conversion of the pixel position coordinates of the fin end plate holes.

[0039] Further, inputting the detection image set into the end plate hole guiding needle recognition model to obtain the coordinates of the guiding needle in the fin end plate holes in the detection image includes:

[0040] S41. Adjusting the size of the detection image and performing normalization processing;

[0041] S42. Extracting the image features of the processed detection image to obtain a feature map;

[0042] S43. Setting the step size of the model, dividing the feature map into r*r grids according to the step size, using the grids as anchor boxes, and obtaining the grid indices of the grid cells;

[0043] S44. Set the prediction of w bounding boxes for each grid cell to obtain r*r*w bounding boxes. Set a threshold and remove redundant bounding boxes through non-maximum suppression. The remaining bounding boxes are the obtained fin hole guiding pin prediction boxes. The center point coordinates of the fin hole guiding pin prediction box are shown in formulas (11) and (12).

[0044] x ct =(σ(t x )·w a )+x a (11)

[0045] y ct =(σ(t y )·h a )+y a (12)

[0046] Among them, (x ct ,y ct ) are the center point coordinates of the fin hole guiding pin prediction box, t x ,t y represent the offset of the center point of a certain fin hole guiding pin prediction box relative to the center point of the corresponding anchor box; w a ,h a represent the width and height of the corresponding anchor box; (x a ,y a ) are the coordinates of the center point of the corresponding anchor box; σ(x) is the Sigmoid function;

[0047] Calculate the guiding pin pixel position coordinates corresponding to a certain fin hole guiding pin prediction box in the image according to the stride of the end plate hole recognition model and the obtained grid index, as shown in formulas (13) and (14).

[0048] X ct =x ct ·s·g x (13)

[0049] X ct =y ct ·s·g y (14)

[0050] Among them, the pixel position coordinates of the guiding pin in the verification image are O′(X ct ,Y ct ), s is the stride, g x and g y are the grid indices;

[0051] Obtain the pixel position coordinates of all guiding pins on the verification image as O′ N (X ctN ,YctN ), N = 1, 2, 3, …, n.

[0052] Further, the coordinates of the guide pins in the fin end plate holes are compared and calculated with the pixel position coordinates of the corresponding fin end plate holes to obtain the offset value of the guide pins relative to the pixel position coordinates of the fin end plate holes. According to the offset value and the physical coordinates of the corresponding fin end plate holes, the physical coordinates of the guide pins in the fin end plate holes are obtained finally, including:

[0053] S51. Set a distance threshold, and set the pixel position coordinates of all guide pins in the detected image as O′ N (X′ N , Y′ N ), N = 1, 2, 3, …, n, and traverse them respectively with the pixel position coordinates of all fin end plate holes on the image to calculate the distance between the pixel position coordinates of the guide pins and the pixel position coordinates of the fin end plate holes, and obtain the result array S less than the set threshold, as shown in formula (15):

[0054]

[0055] where O N is the pixel position coordinate of the fin end plate hole in the detected image, d is the distance threshold, and K is the conversion coefficient for converting pixel position coordinates to physical coordinates; when the calculated distance is less than the actual normal distance between two holes, there is one and only one coordinate element in the array S, that is, the pixel position coordinate of the fin end plate hole corresponding to the pixel position coordinate of the guide pin in the fin end plate hole is found;

[0056] S52. Subtract the pixel position coordinate of the guide pin in the fin end plate hole from the pixel position coordinate of the corresponding fin end plate hole, and multiply by the conversion coefficient to obtain the offset value of the actual guide pin pixel position coordinate relative to the pixel position coordinate of the corresponding fin end plate hole, as shown in formulas (16) and (17):

[0057] X DN = K·(X ctN - X N ) (16)

[0058] Y DN = K·(Y ctN - Y N ) (17)

[0059] where X D and Y D are the offset values of the abscissa and ordinate respectively;

[0060] S53. According to the offset values X D , Y DCalculate the actual physical coordinates of the guiding needle in the fin end plate hole based on the physical coordinates of the corresponding fin end plate hole, as shown in Formulas (18) and (19).

[0061] X VN = X UN + X D (18)

[0062] Y VN = Y UN + Y D (19)

[0063] The actual physical position coordinates of the guiding needle in the final fin end plate hole are O VN (X VN , Y VN ).

[0064] Beneficial effects: The present invention provides a method for identifying the fin end of a heat exchanger in the refrigeration field. First, through the positioning of two Mark points, the pixel coordinates and physical coordinates of all the holes on the fin are obtained. By introducing the YOLO object detection network to train the image, after deep learning training, all the guiding needle and hole coordinates on the fin can be output at one time, with fast speed, convenient calculation, higher calculation result accuracy, and more stable identification effect. Then, by comparing and calculating between the guiding needle coordinates and the pixel coordinates of the holes, the final more accurate offset value of the guiding needle relative to the hole is obtained, and further the actual physical coordinates of the guiding needle are obtained, improving the accuracy of automatically inserting the heat exchange tube and meeting the production requirements of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order 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 the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0066] Figure 1 is a flowchart of a method for identifying the fin end of a heat exchanger in the refrigeration field provided by the present invention;

[0067] Figure 2 is the original image of the air conditioner fin end plate;

[0068] Figure 3 is the effect diagram of the spliced image of the fin end plate;

[0069] Figure 4 is the calculation diagram of the Mark point positioning reference of the present invention;

[0070] Figure 5 is the structural diagram of the YOLO algorithm;

[0071] Figure 6 This is the correlation result graph for the deep learning training of the present invention;

[0072] Figure 7 This is the secondary positioning result graph for the deep learning using YOLO in the present invention;

[0073] Figure 8 Partial enlarged view of the image of the fin end plate hole recognized and positioned by the present invention. Detailed implementation manners

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0075] This embodiment provides a method for identifying the fin end of a heat exchanger in the refrigeration field, as Figure 1 shown, including:

[0076] S1: Obtain an image set of the fin end of an air conditioner, preprocess the image of the fin end of the air conditioner to obtain a detection image set, and the detection image includes a fin end plate hole and a guiding pin in the fin end plate hole;

[0077] S2: Use a template matching tool to obtain the pixel position coordinates of the Mark point in the fin end plate hole in the detection image set, obtain the pixel position coordinates of the fin end plate hole according to the pixel position coordinates of the Mark point, and further obtain the physical coordinates of the fin end plate hole according to the pixel position coordinates of the fin end plate hole;

[0078] S3: Mark the position of the guiding pin in the fin end plate hole in the detection image set to obtain a marked image set, input the marked image set into a YOLO neural network for training to obtain a fin end plate hole guiding pin recognition model;

[0079] S4: Input the detection image set into the fin end plate hole guiding pin recognition model to obtain the guiding pin coordinates in the fin end plate hole in the detection image;

[0080] S5: Compare and calculate the guiding pin coordinates in the fin end plate hole with the pixel position coordinates of the corresponding fin end plate hole to obtain the offset value of the guiding pin relative to the pixel position coordinates of the fin end plate hole, and obtain the physical coordinates of the guiding pin in the fin end plate hole in the final actual situation according to the offset value and the physical coordinates of the corresponding fin end plate hole.

[0081] Specifically, first, an image set of the end part of the air conditioner fin is obtained, and the images of the end part of the air conditioner fin are preprocessed to obtain a set of detection images. The detection images include fin end plate holes and guide pins in the fin end plate holes. Images of the fin holes are obtained and processed, which solves the problem that each guide pin hole in the fin picture is too fixedly concentrated and the data is unbalanced, and also improves the training and learning efficiency to a certain extent. Secondly, a template matching tool is used to obtain the pixel position coordinates of the Mark points in the fin end plate holes in the set of detection images, and the pixel position coordinates of the fin end plate holes are obtained according to the pixel position coordinates of the Mark points. Furthermore, the physical coordinates of the fin end plate holes are obtained according to the pixel position coordinates. Using the Mark point coordinates can provide a more accurate reference position, improve the positioning accuracy, and reduce errors.

[0082] Thirdly, the positions of the guide pins in the fin end plate holes in the set of detection images are marked to obtain a set of marked images. The set of marked images is input into the YOLO neural network for training to obtain a recognition model for the guide pins in the end plate holes. Then, the set of detection images is input into the recognition model for the guide pins in the end plate holes to obtain the coordinates of the guide pins in the fin end plate holes in the detection images. Using the YOLO neural network to perform network feature learning on each marked image converts the object detection task into a regression problem, greatly accelerating the detection speed and enabling the object to be very accurately positioned.

[0083] Finally, the coordinates of the guide pins in the fin end plate holes are compared and calculated with the pixel position coordinates of the corresponding fin end plate holes to obtain the offset value of the pixel position coordinates of the guide pins relative to the fin end plate holes. According to the offset value and the physical coordinates of the corresponding fin end plate holes, the physical coordinates of the guide pins in the fin end plate holes in the final actual situation are obtained. Obtaining the offset value and the actual object coordinates can perform a machine guiding compensation operation with higher precision and judge defects such as the deformation amount of the fins. The equipment knows the reference coordinates of the fin end plate holes. After obtaining the offset value between the fin holes and the guide pins as described above, the reference coordinates of the holes are added with the offset value to finally obtain the actual coordinates of the guide pins on the fin end plate. In the subsequent process, the air conditioner copper tube is precisely connected and matched with the guide pins at the end of the fin to perform a more accurate interpolation operation, improving the accuracy of inserting the heat exchange tube into the fin holes; and the deformation size of the fin end can be measured according to the size of the actual offset value between the fin hole reference and the guide pin. The larger the offset value indicates the greater the misalignment between the position of the guide pin and the fin hole or the deformation of the fin end itself. Defective products caused by the above reasons can be screened out by controlling the offset value.

[0084] In a specific embodiment, the solution for obtaining an image set of the end part of the air conditioner fin and preprocessing the images of the end part of the air conditioner fin to obtain a set of detection images, where the detection images include fin end plate holes and guide pins in the fin end plate holes is as follows:

[0085] Use a 20 - megapixel grayscale camera to repeatedly take pictures of different air - conditioner fin end - plates to collect picture samples. As Figure 2 shown, since the fin guiding pins are relatively concentrated in the image, using the Mosaic data augmentation method, splice the fin end - plate hole pictures in a 2×2 manner. The number of end - plate holes is between 20 - 30, and there are multiple end - plate holes in each row. The positions of the end - plate holes are consistent with those of the fin holes. Behind the end - plate holes are the fin holes. The guiding pins are located inside the fin holes and are concentrically aligned with the fin holes. The guiding pins are used for accurate positioning and guiding to ensure that the copper tube can pass through the fin holes smoothly. The effect of the spliced image is as Figure 3 shown.

[0086] In this solution, obtaining and processing the images of the fin holes solves the problem of the over - fixed concentration of each guiding - pin hole in the fin pictures and the data imbalance, and also improves the training and learning efficiency to a certain extent.

[0087] In a specific embodiment, the scheme of using a template - matching tool to obtain the pixel position coordinates of the Mark points in the fin end - plate holes in the detection image set, obtaining the pixel position coordinates of the fin end - plate holes according to the pixel position coordinates of the Mark points, and then obtaining the physical coordinates of the fin end - plate holes according to the pixel position coordinates of the fin end - plate holes is as follows:

[0088] S21. Randomly select a detection image as a reference image, and use a template - matching tool to select two end - plate holes closest to the image center and at the corners on the reference image as two reference Mark points. In this embodiment, the corner is the upper - right corner, as Figure 4 shown:

[0089] In this embodiment, through the template - matching tool matchTemplate, use the CCOEFF_NORMED standard normalization algorithm for matching. Select the maximum value from the obtained data as the best - matching result, and then obtain the final image - matching point coordinates according to the template's own size data. In this way, obtain the pixel position coordinates of the reference Mark points in sequence, which are P1(x1,y1) and P2(x2,y2) respectively, where P1 is used as the origin;

[0090] S22. Randomly select a detection image, and re - obtain the pixel position coordinates of the two reference Mark points on this detection image, which are P′ N (x′ N ,y′ N ) and P′ N+1 (x′ N+1 ,y′ N+1 ), N ∈ 1,2,3…n, and P′ N is used as the origin;

[0091] S23. Use the affine transformation algorithm to calculate the pixel position coordinates of each fin end plate hole on the detected image according to the coordinates of two sets of Mark points:

[0092] S231: Calculate the angle between the pixel position coordinates of two reference Mark points on the detected image and the pixel position coordinates of two reference Mark points on the reference image, as shown in formula (20).

[0093]

[0094] where tanθ represents the tan value of the angle between the pixel position coordinates of two reference Mark points on the detected image and the two reference Mark points coordinates on the reference image; θ is the angle between the pixel position coordinates of two reference Mark points on the detected image and the pixel position coordinates of two reference Mark points on the reference image.

[0095] S232: Calculate the translation vector and unit vector through two points (used for scaling, if scaling is not required, the unit vector is (1,0) or (0,1)). However, usually, for simple translation and rotation (without scaling), we can directly calculate the translation vector through the point coordinate difference, that is, obtain the offset of the origin of the detected image relative to the origin of the reference image, as shown in formulas (21) and (22).

[0096] Δx N =x′ N -x1 (21)

[0097] Δy N =y′ N -y1 (22)

[0098] where, Δx N represents the translation amount of the abscissa, and Δy N represents the translation amount of the ordinate.

[0099] S233: Construct the affine transformation matrix of the fin end plate holes on the detected image according to the calculated angle and offset, as shown in formula (23).

[0100]

[0101] S234: Obtain the transformed pixel position coordinates of a certain fin end plate hole in the reference image on the detected image according to the affine transformation matrix, as shown in formula (24).

[0102]

[0103] where, O0(X0,Y0) represents the pixel position coordinates of the fin end plate hole in the reference image, O N (X N ,YN ) represents the pixel position coordinates after transformation of the fin end plate holes in the reference image on the detection image;

[0104] X N and Y N As shown in Formula (25) and Formula (26),

[0105] X N = X0 cosθ - Y0 sinθ + Δx N (25)

[0106] Y N = X0 sinθ + Y0 cosθ + Δy N (26)

[0107] S24. Calculate the conversion coefficient according to the physical distance between two adjacent fin end plate holes in the reference image and the pixel position coordinates of two reference Mark points. The conversion coefficient is used to convert the pixel position coordinates into physical coordinates, and the conversion coefficient is as shown in Formula (27),

[0108]

[0109] where K is the conversion coefficient and L is the physical distance between two adjacent fin end plate holes; in this embodiment, L is 20 mm;

[0110] S25. According to the conversion coefficient and the origin P1(x1, y1) of the reference image, convert the pixel position coordinates O of the fin end plate holes in the detection image N (X N , Y N ) into physical coordinates, as shown in Formula (28) and Formula (29),

[0111] X UN = (X N - x1)·K (28)

[0112] Y UN = (Y N - y1)·K (29)

[0113] U N (X UN , Y UN ) represents the physical coordinates after conversion of the pixel position coordinates of the fin end plate holes.

[0114] In this solution, Mark point coordinates are used. By identifying two Mark points, the pixel coordinate set of all fin holes can be quickly obtained, which is R = {O1, O2, O3.....O n}, the point closest to the center of the image, i.e., the middle Mark point, is taken as the origin. According to the actual fin end plate coordinate hole pitch of 20 mm, the physical coordinate set W = {U1, U2, U3.....U n};

[0115] Using the Mark point coordinates can provide a more accurate reference position, improve the positioning accuracy, and reduce errors.

[0116] In a specific embodiment, the method for annotating the positions of the guide pins in the fin end plate holes in the detection image set to obtain an annotated image set, and inputting the annotated image set into a YOLO neural network for training to obtain a fin end plate hole guide pin recognition model is as follows:

[0117] S31. According to the coordinates, use a standard tool to mark the positions of the guide pins in each fin hole area, and divide the annotated image set into a training set and a validation set with a ratio of 7:3;

[0118] S32. Input the training set into a YOLO neural network for training to obtain a fin end plate hole guide pin recognition model:

[0119] The model of the YOLO neural network is as Figure 5 shown. Through the YOLO neural network, the end plate holes in the fin image are classified, and the fins in different forms after annotation are subjected to multiple rounds of repetitive training to update the weights of each layer of the neural network. When training, the SGD optimizer is selected, the training batch size is 16, and after about 150 rounds of training, the precision curve and the mAP_0.5 curve approach 1. The number of training Epochs is set to 200, and after 200 rounds of training, a depth model for recognizing end plate holes is obtained. The training results are as Figure 6 shown. It can be seen from the figure that the Los loss function approaches 0, the training results are saturated, and the Precision / Recall curve approaches 1. Therefore, it has excellent detection performance.

[0120] YOLO (You Only Look Once) is an efficient object detection algorithm. Its core idea is to transform the object detection problem into a regression problem, and the object detection can be completed through a single forward propagation. In this embodiment, the YOLO neural network is used to perform network feature learning on each annotated image, and the object detection task is converted into a regression problem, which greatly speeds up the detection speed, enabling YOLO to process 60 grayscale images of 20 million pixels per second. Moreover, since each network predicts the target window using a local area of the image, the target can be very accurately located.

[0121] In a specific embodiment, the solution for inputting the detection image set into the end plate hole guiding needle recognition model to obtain the coordinates of the guiding needle in the fin end plate hole in the detection image is as follows:

[0122] S41. Adjust the size of the detection image and perform normalization processing;

[0123] S42. Extract the image features of the processed detection image to obtain a feature map;

[0124] S43. Set the stride of the model, divide the feature map into r*r grids according to the stride, use the grids as anchor boxes, and obtain the grid indices of the grid cells:

[0125] Adopt YOLO target depth object detection to divide the above labeled picture into 7*7 (assuming S = 7) grid cells. The grid cells are regarded as anchor boxes. For each anchor box, set to predict 2 bounding boxes (including the confidence that each bounding box is a target and the probabilities of each bounding box area in multiple categories), and 7*7*2 target windows can be predicted;

[0126] S44. Set a threshold, and remove redundant bounding boxes through non-maximum suppression. The remaining bounding boxes are the predicted boxes of the fin hole guiding needle obtained: The center point coordinates of the predicted box of the fin hole guiding needle are shown in formulas (30) and (31) as follows,

[0127] x ct =(σ(t x )·w a )+x a (30)

[0128] y ct =(σ(t y )·h a )+y a (31)

[0129] where, (x ct ,y ct ) are the center point coordinates of the predicted box of the fin hole guiding needle, t x ,t y represent the offset of the center point of a certain predicted box of the fin hole guiding needle relative to the center point of the corresponding anchor box; w a ,h a represent the width and height of the corresponding anchor box; (x a ,y a ) are the center point coordinates of the corresponding anchor box; σ(x) is the Sigmoid function;

[0130] Calculate the guiding needle pixel position coordinates corresponding to a certain fin hole guiding needle prediction box in the image according to the step size of the end plate hole recognition model and the obtained grid index, as shown in formulas (32) and (33).

[0131] X ct = x ct ·s·g x (32)

[0132] Y ct = y ct ·s·g y (33)

[0133] Among them, the pixel coordinates of the guiding needle in the image are O′(X ct , Y ct ), s is the step size, g x and g y are the grid indices;

[0134] Obtain the pixel position coordinates of all guiding needles on the verification image as O′ N (X ctN , Y ctN ), N = 1, 2, 3, …, n, that is, the position set of all guiding needles on the detection image is T = {O′1, O′2, O′3…O n}.

[0135] Perform secondary positioning on the detection image to perform fin recognition, as Figure 7 and Figure 8 shown. The red box part in the figure is the recognized hole position, and the numbers represent the recognition accuracy. It can be seen from the figure that applying YOLO for target recognition can recognize the actual fin end plate hole positions with relatively high accuracy.

[0136] In a specific embodiment, the scheme of comparing and calculating the coordinates of the guiding needle in the fin end plate hole with the pixel position coordinates of the corresponding fin end plate hole to obtain the offset value of the guiding needle relative to the pixel position coordinates of the fin end plate hole, and obtaining the physical coordinates of the guiding needle in the actual fin end plate hole according to the offset value and the physical coordinates of the corresponding fin end plate hole is as follows:

[0137] S51. Set a distance threshold, and set the pixel position coordinates of all guiding needles in the detection image as O′ N (X′ N , Y′ N ), N = 1, 2, 3, …, n, and traverse them respectively with the pixel position coordinates of all fin end plate holes on this image to calculate the distance between the guiding needle pixel position coordinates and the pixel position coordinates of the fin end plate hole, and obtain the result array S with values less than the set threshold, as shown in formula (34).

[0138]

[0139] Among them, d is the distance threshold; the actual pitch of each fin hole is 20 mm, and in this embodiment, d is set to 5 mm; O N is the pixel position coordinate of the fin end plate hole in the detected image, d is the distance threshold, and K is the conversion coefficient for converting the pixel position coordinate to the physical coordinate; when the calculated distance is less than the actual normal hole pitch, there is exactly one coordinate element in the array S, that is, the pixel position coordinate of the fin end plate hole corresponding to the pixel position coordinate of the guide pin of the fin end plate hole is found;

[0140] S52. Subtract the pixel position coordinate of the guide pin of the fin end plate hole from the corresponding pixel position coordinate of the fin end plate hole, and multiply by the conversion coefficient to obtain the offset value of the actual pixel position coordinate of the guide pin relative to the corresponding pixel position coordinate of the fin end plate hole, as shown in formulas (35) and (36),

[0141] X DN = K·(X ctN - X N ) (35)

[0142] Y DN = K·(Y ctN - Y N ) (36)

[0143] Among them, X D and Y D are the offset values of the abscissa and ordinate respectively;

[0144] S53. Calculate the actual physical coordinates of the guide pin in the fin end plate hole according to the offset values X D , Y D and the corresponding physical coordinates of the fin end plate hole, as shown in formulas (37) and (38),

[0145] X VN = X UN + X D (37)

[0146] Y VN = Y UN + Y D (38)

[0147] The actual physical position coordinates of the guide pin in the fin end plate hole finally are O VN (X VN , Y VN ).

[0148] The physical position coordinates O of the guide pin obtained through depth recognitionVN (X VN , Y VN ), and the offset values X DN , Y DN , higher-precision machine-guided compensation operations and defect judgments such as fin deformation can be carried out. The equipment knows the reference coordinates of the fin end plate holes. After obtaining the offset values between the fin holes and the guide pins as described above, the hole reference coordinates are added with the offset values to finally obtain the actual coordinates of the fin end plate guide pins. In the subsequent process, the air-conditioning copper tube is precisely connected and matched with the fin end guide pins to perform more accurate interpolation operations, improving the accuracy of inserting the heat exchange tube into the fin holes; and the magnitude of the deformation of the fin end can be measured according to the size of the actual offset value between the fin hole reference and the guide pin. The larger the offset value, the greater the misalignment between the position of the guide pin and the fin hole or the greater the deformation of the fin end itself. Defective products caused by the above reasons can be screened out through the control of the offset value.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying fin ends of a heat exchanger in the field of refrigeration, characterized in that: include: S1: acquiring an image set of an air conditioner fin end, and preprocessing the image set of the air conditioner fin end to obtain a detection image set, wherein the detection image includes a fin end plate hole and a guide needle in the fin end plate hole; S2: using a template matching tool to obtain pixel position coordinates of a Mark point in the fin end plate hole in the detection image set, obtaining pixel position coordinates of the fin end plate hole according to the pixel position coordinates of the Mark point, and further obtaining physical coordinates of the fin end plate hole according to the pixel position coordinates of the fin end plate hole; S3: marking the position of the guide needle in the fin end plate hole in the detection image set to obtain a marked image set, and inputting the marked image set into a YOLO neural network for training to obtain an end plate hole guide needle recognition model; S4: inputting the detection image set into the end plate hole guide needle recognition model to obtain the guide needle coordinates in the fin end plate hole in the detection image; S5: Compare and calculate the coordinates of the guide needle in the fin end plate hole with the pixel position coordinates of the corresponding fin end plate hole to obtain the offset value of the guide needle relative to the pixel position coordinates of the fin end plate hole, and obtain the final actual physical coordinates of the guide needle in the fin end plate hole based on the offset value and the physical coordinates of the corresponding fin end plate hole.

2. A method for identifying fin ends of a heat exchanger in the refrigeration field according to claim 1, characterized in that: The pixel position coordinates of the Mark point in the fin end plate hole in the detection image set are obtained using a template matching tool, including: S21, randomly select a detection image as a reference image, use a template matching tool to select two end plate holes closest to the center and the corner of the image on the reference image as two reference Mark points, the pixel position coordinates of the two reference Mark points are P1 (x1, y1) and P2 (x2, y2), respectively, where P1 is the origin; S22, randomly select another detection image, and re-obtain the pixel position coordinates of the two reference Mark points on the detection image, which are P′ N (x′ N ,y′ N ) and P′ N+1 (x′ N+1 ,y′ N+1 ),N∈1,2,3…n,P′ N as the origin; S23, using an affine transformation algorithm to calculate the pixel position coordinates of each fin end plate hole on the detection image according to the two sets of Mark point coordinates.

3. A method for identifying fin ends of a heat exchanger in the refrigeration field according to claim 2, characterized in that: Using an affine transformation algorithm to calculate the pixel position coordinates of each fin end plate hole on the detection image according to the two sets of Mark point coordinates, including: S231: Calculate the angle between the pixel position coordinates of the two reference Mark points of the detection image and the pixel position coordinates of the two reference Mark points of the reference image, as shown in formula (1): Wherein, tanθ represents the tan value of the angle between the pixel position coordinates of the two reference Mark points of the detection image and the pixel position coordinates of the two reference Mark points of the reference image; θ represents the angle between the pixel position coordinates of the two reference Mark points of the detection image and the pixel position coordinates of the two reference Mark points of the reference image; S232: Obtain the offset of the origin of the detection image relative to the origin of the reference image, as shown in formulas (2) and (3). Δx N =x′ n -x1 (2) Δy N =y′ N -y1 (3) Where Δx N Indicates the offset of the horizontal axis, Δy N Indicates the offset of the vertical axis; S233: Construct an affine transformation matrix of the fin end plate hole of the detection image according to the calculated angle and offset, as shown in formula (4): S234: Obtain the pixel position coordinates of a fin end plate hole in the reference image after transformation on the detection image according to the affine transformation matrix, as shown in formula (5): Among them, O0(X0,Y0) represents the pixel position coordinates of the fin end plate hole in the reference image, O N (X N ,Y N ) represents the pixel position coordinates of the fin end plate hole in the reference image after transformation on the detection image; X N and Y N As shown in formula (6) and formula (7), X N =X0cosθ-Y0sinθ+Δx N (6) Y N =X0sinθ+Y0cosθ+Δy N (7)。 4. A method for identifying fin ends of a heat exchanger in the refrigeration field according to claim 3, characterized in that: Obtaining the physical coordinates of the fin end plate hole according to the pixel position coordinates of the fin end plate hole includes: S24, calculating a conversion coefficient according to the physical spacing between two adjacent fin end plate holes in the reference image and the pixel position coordinates of two reference Mark points in the reference image, wherein the conversion coefficient is used to convert the pixel position coordinates into physical coordinates, and the conversion coefficient is shown in formula (8), Where K is the conversion coefficient, and L is the physical distance between two adjacent fin end plate holes; S25, converting the pixel position coordinates O of the fin end plate hole of the detection image into the pixel coordinates O of the fin end plate hole of the detection image according to the conversion coefficient and the origin P1 (x1, y1) of the reference image. N (X N ,Y N ) is converted into physical coordinates, as shown in formula (9) and formula (10), X UN =(X N -x1)·K (9) T UM =(Y N -y1)·K (10) U N (X UN , Y UN ) represents the physical coordinates after the pixel position coordinates of the fin end plate hole are converted.

5. A method for identifying fin ends of a heat exchanger in the refrigeration field according to claim 1, characterized in that: Inputting the detection image set into the end plate hole guide needle recognition model to obtain the guide needle coordinates in the fin end plate hole in the detection image, including: S41, adjusting the size of the detection image and performing normalization processing; S42, extracting image features of the processed detection image to obtain a feature map; S43, setting the step size of the model, dividing the feature map into r*r grids according to the step size, using the grids as anchor boxes, and obtaining grid indexes of the grid cells; S44, set each grid unit to predict w frames, obtain r*r*w frames, set a threshold, remove redundant frames by non-maximum suppression method, and the remaining frames are the obtained fin hole guide needle prediction frames. The center point coordinates of the fin hole guide needle prediction frames are shown in formulas (11) and (12). x ct =(σ(t x )·w a )+x a (11) y ct =(σ(t y )·h a )+y a (12) Among them, (x ct ,y ct ) is the center coordinate of the fin hole guide needle prediction frame, t x ,t y represents the offset of the center point of a fin hole guide needle prediction frame relative to the center point of the corresponding anchor point frame; w a ,h a Indicates the width and height of the corresponding anchor box; (x a ,y a ) is the coordinate of the center point of the corresponding anchor box; σ(x) is the Sigmoid function; The pixel position coordinates of the guide needle corresponding to a guide needle prediction frame of a fin hole in the image are calculated according to the step size of the end plate hole recognition model and the obtained grid index, as shown in formulas (13) and (14): X ct =x ct ·s·g x (13) X ct =y ct ·s·g y (14) The pixel position coordinates of the guide needle in the verification image are O′(X ct ,Y ct ), s is the step size, g x and g y is the grid index; The pixel position coordinates of all guide needles on the verification image are obtained by formulas (11)-(14) as O′ N (X ctN ,Y ctN ),N=1,2,3,…,n.

6. A method for identifying fin ends of a heat exchanger in the refrigeration field according to claim 5, characterized in that: Compare and calculate the coordinates of the guide needle in the fin end plate hole with the pixel position coordinates of the corresponding fin end plate hole to obtain the offset value of the guide needle relative to the pixel position coordinates of the fin end plate hole, and obtain the final actual physical coordinates of the guide needle in the fin end plate hole according to the offset value and the physical coordinates of the corresponding fin end plate hole, including: S51, set the distance threshold, and set the pixel position coordinates of all guide needles in the detection image to O' N (X′ N ,Y′ N ), N = 1, 2, 3, ..., n, respectively traverse the pixel position coordinates of all fin end plate holes on the image, calculate the distance between the pixel position coordinates of the guide needle and the pixel position coordinates of the fin end plate hole, and obtain the result array S that is less than the set threshold, as shown in formula (15), Among them, O N To detect the pixel position coordinates of the fin end plate hole in the image, d is the distance threshold, and K is the conversion coefficient of the pixel position coordinate to the physical coordinate; when the calculated distance is less than the actual normal distance between the two holes, there is only one coordinate element in the array S, that is, the pixel position coordinates of the fin end plate hole corresponding to the pixel position coordinates of the guide needle of the fin end plate hole are found; S52, subtract the pixel position coordinates of the guide needle of the fin end plate hole from the pixel position coordinates of the corresponding fin end plate hole, and multiply by the conversion coefficient to obtain the offset value of the actual guide needle pixel position coordinate relative to the corresponding fin end plate hole pixel position coordinate, as shown in formulas (16) and (17), X DN =K·(X ctN -X N ) (16) AND DN =K·(Y ctN -AND N ) (17) Among them, X D and Y D are the offset values ​​of the horizontal and vertical axes respectively; S53, according to the offset value X D ,Y D The actual physical coordinates of the guide needle in the fin end plate hole are calculated by using the physical coordinates of the corresponding fin end plate hole, as shown in formulas (18) and (19), X VN =X UN +X D (18) AND VN =And UN +Y D (19) The actual physical position coordinates of the guide needle in the fin end plate hole are finally VN (X VN ,Y VN ).

Images