A method, system, device, and medium for adjusting bake temperature of a painting line

Abstract

The application discloses a coating production line baking temperature adjusting method, system, device and medium, and belongs to the technical field of intelligent manufacturing technology of engineering machinery. The coating production line baking temperature adjusting method comprises the following steps: acquiring a two-dimensional image of a workpiece hanging area of a coating production line, generating a workpiece category detection result of each workpiece in the two-dimensional image by using a target detection algorithm model; acquiring a three-dimensional point cloud of the workpiece hanging area, and generating a corresponding size type detection result for each workpiece in the three-dimensional point cloud; determining the total mass of the workpieces in the batch according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area; determining a corresponding target baking temperature according to the total mass of the workpieces in the batch, and controlling the coating production line to perform a corresponding baking operation according to the target baking temperature. The application can realize high-precision and automatic baking temperature adjustment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing of engineering machinery, and in particular relates to a baking temperature adjusting method, system, device and medium for a coating production line. BACKGROUND

[0002] In the coating production line, there is a process of baking workpieces. In the related art, the baking temperature of the coating production line is usually set according to manual experience; however, since the manual setting of the temperature cannot be standardized, energy waste and factory emissions increase.

[0003] Therefore, how to realize high-precision and automatic baking temperature adjustment is a technical problem that needs to be solved by those skilled in the art at present. SUMMARY

[0004] The purpose of the present application is to provide a baking temperature adjusting method, system, device and medium for a coating production line, which can realize high-precision and automatic baking temperature adjustment.

[0005] To solve the above technical problems, the present application provides a baking temperature adjusting method for a coating production line, comprising:

[0006] Obtaining a two-dimensional image of a workpiece hanging area of a coating production line, and generating a workpiece category detection result of each workpiece in the two-dimensional image by using a target detection algorithm model;

[0007] Obtaining a three-dimensional point cloud of the workpiece hanging area, and generating a corresponding size type detection result for each workpiece in the three-dimensional point cloud;

[0008] Determining the total mass of the workpieces in the batch according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area;

[0009] Determining a corresponding target baking temperature according to the total mass of the workpieces in the batch, and controlling the coating production line to perform a corresponding baking operation according to the target baking temperature.

[0010] Optionally, determining the total mass of the workpieces in the batch according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area comprises:

[0011] Selecting a target workpiece from the workpiece hanging area;

[0012] Querying a size type corresponding to the workpiece category detection result of the target workpiece in a database to obtain a size type query result;

[0013] Judging whether the size type detection result of the target workpiece is the same as the size type query result;

[0014] If yes, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type detection result;

[0015] If no, determining a two-dimensional detection confidence and a three-dimensional detection confidence of the target workpiece; wherein the two-dimensional detection confidence is a confidence of the workpiece category detection result, and the three-dimensional detection confidence is a confidence of the size type detection result;

[0016] If the two-dimensional detection confidence is greater than the three-dimensional detection confidence, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result;

[0017] If the two-dimensional detection confidence is less than the three-dimensional detection confidence, determining the workpiece quality of the target workpiece according to the size type detection result;

[0018] If the two-dimensional detection confidence is equal to the three-dimensional detection confidence, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type detection result;

[0019] Determining whether all workpieces in the workpiece hanging area are selected; if yes, calculating the total quality of the batch of workpieces according to the workpiece qualities of all workpieces in the workpiece hanging area; if no, entering the step of selecting a target workpiece from the workpiece hanging area.

[0020] Optionally, after the target workpiece is selected from the workpiece hanging area, the method further comprises:

[0021] If the target workpiece has a corresponding workpiece category detection result and does not have a corresponding size type detection result, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result;

[0022] If the target workpiece has a corresponding size type detection result and does not have a corresponding workpiece category detection result, determining the workpiece quality of the target workpiece according to the size type detection result.

[0023] Optionally, determining the workpiece quality of the target workpiece according to the size type detection result comprises:

[0024] Querying alternative workpieces matching the size type detection result in a database;

[0025] Setting the average quality of all the alternative workpieces as the workpiece quality of the target workpiece.

[0026] Optionally, before generating the workpiece category detection result of each workpiece in the two-dimensional image by using the target detection algorithm model, the method further comprises:

[0027] obtaining training samples from the data set; wherein each of the training samples comprises an original sample, a positive sample and a negative sample, the original sample being an image containing a workpiece, the positive sample being an image within a bounding box of the workpiece in the original sample, and the negative sample being an image outside the bounding box of the workpiece in the original sample;

[0028] training the target detection algorithm model by using the training samples;

[0029] The target detection algorithm model comprises a skeleton network, a neck network, a detection head, a contrast learning module and a loss calculation module. The contrast learning module comprises a global average pooling layer, a multilayer perceptron and a spatial attention layer. The contrast learning module is configured to convert a feature map input by the skeleton network into a positive sample global enhanced feature map.

[0030] Optionally, the process in which the contrast learning module converts the feature map input by the skeleton network into the positive sample global enhanced feature map comprises:

[0031] receiving an original sample feature map, a positive sample feature map and a negative sample feature map output by the skeleton network;

[0032] inputting the original sample feature map, the positive sample feature map and the negative sample feature map into the global average pooling layer to obtain global channel weights; wherein the global channel weights comprise an original sample global channel weight, a positive sample global channel weight and a negative sample global channel weight;

[0033] point-multiplying the original sample feature map and the original sample global channel weight to obtain a global enhanced feature map;

[0034] inputting the original sample global channel weight, the positive sample global channel weight and the negative sample global channel weight into the multilayer perceptron to obtain a processing result;

[0035] inputting the original sample feature map, the positive sample feature map and the negative sample feature map into the spatial attention layer to obtain spatial local feature weights;

[0036] point-multiplying the original sample feature map, the positive sample feature map and the negative sample feature map respectively and corresponding spatial local feature weights to obtain local enhanced feature maps; wherein the local enhanced feature maps comprise an original sample local enhanced feature map, a positive sample local enhanced feature map and a negative sample local enhanced feature map;

[0037] performing an element subtraction operation on the original sample local enhanced feature map and the negative sample local enhanced feature map to obtain a local information attention feature;

[0038] According to the processing result and the local enhanced feature map, a contrast loss function value is calculated, and according to the positive sample local enhanced feature map and the local information attention feature, an approximation loss function value is calculated, so as to optimize the parameters of the contrast learning module by using the contrast loss function value and the approximation loss function value.

[0039] An element addition operation is performed on the global enhanced feature map and the local information attention feature to obtain a positive sample global enhanced feature map, wherein the positive sample global enhanced feature map is used as input to the neck network.

[0040] Optionally, a corresponding size type detection result is generated for each workpiece in the three-dimensional point cloud, including:

[0041] A point cloud filtering and segmentation operation is performed on the three-dimensional point cloud to obtain a workpiece point cloud of each workpiece.

[0042] According to the workpiece point cloud, a point cloud plane equation of each workpiece is calculated, and an inclination angle of the point cloud plane equation is determined.

[0043] According to the point cloud plane equation, a minimum bounding box of the workpiece point cloud is determined, and a volume of the workpiece point cloud is calculated according to the minimum bounding box.

[0044] According to the area of the workpiece point cloud, the volume of the workpiece point cloud and the inclination angle of the point cloud plane equation, a size type detection result of each workpiece is determined.

[0045] Optionally, the coating production line is provided with a workpiece hanging rack, a three-dimensional laser camera and a plurality of two-dimensional cameras, the workpiece hanging area is an area below the workpiece hanging rack, all the two-dimensional cameras are in the same plane, and the shooting direction of the three-dimensional laser camera is perpendicular to the shooting direction of each two-dimensional camera.

[0046] Correspondingly, acquiring a two-dimensional image of a workpiece hanging area of a coating production line includes:

[0047] Capturing local images of the workpiece hanging area of the coating production line from multiple directions by using all the two-dimensional cameras, and splicing the local images to obtain the two-dimensional image.

[0048] Correspondingly, acquiring a three-dimensional point cloud of the workpiece hanging area includes:

[0049] Controlling the three-dimensional laser camera to move to a preset position in sequence, and acquiring a local point cloud of the workpiece hanging area at the preset position.

[0050] Splicing the local point cloud to obtain the three-dimensional point cloud of the workpiece hanging area.

[0051] Optionally, before determining the corresponding target baking temperature according to the total mass of the current batch of workpieces, further comprising:

[0052] determining a mapping relationship between the total mass of workpieces and the baking temperature according to historical data, establishing a temperature prediction model according to the mapping relationship, and optimizing the temperature prediction model using a gradient descent algorithm;

[0053] Correspondingly, determining the corresponding target baking temperature according to the total mass of the current batch of workpieces, and controlling the coating production line to perform corresponding baking operations according to the target baking temperature, comprising:

[0054] inputting the total mass of the current batch of workpieces into the temperature prediction model to obtain the target baking temperature;

[0055] Adjusting the natural gas regulating valve opening according to the target baking temperature, so as to control the hot air machine to perform baking operations on the workpieces in the coating production line according to the target baking temperature.

[0056] The present application also provides a baking temperature adjustment system for a coating production line, which comprises:

[0057] a two-dimensional image processing module for acquiring a two-dimensional image of a workpiece hanging area of a coating production line, and generating a workpiece category detection result for each workpiece in the two-dimensional image using a target detection algorithm model;

[0058] a three-dimensional point cloud processing module for acquiring a three-dimensional point cloud of the workpiece hanging area, and generating a corresponding size type detection result for each workpiece in the three-dimensional point cloud;

[0059] a mass calculation module for determining the total mass of the current batch of workpieces according to the workpiece category detection result and / or size type detection result of each workpiece in the workpiece hanging area;

[0060] a temperature control module for determining the corresponding target baking temperature according to the total mass of the current batch of workpieces, and controlling the coating production line to perform corresponding baking operations according to the target baking temperature.

[0061] The present application also provides a storage medium having a computer program stored thereon, wherein the computer program performs the steps of the above-mentioned baking temperature adjustment method for a coating production line when executed.

[0062] The present application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned baking temperature adjustment method for a coating production line when calling the computer program in the memory.

[0063] The application discloses a baking temperature adjusting method of a coating production line. The method obtains a two-dimensional image and a three-dimensional point cloud of a workpiece hanging area, determines the category of the workpiece in the workpiece hanging area according to the two-dimensional image, and identifies the size of the workpiece in the workpiece hanging area according to the three-dimensional point cloud. The quality of the workpiece is related to the category and size of the workpiece. On the basis of obtaining the workpiece category detection result and the size type detection result, the application determines the total mass of the batch of workpieces, and then determines the corresponding target baking temperature according to the total mass of the batch of workpieces, so as to perform the corresponding baking operation according to the target baking temperature. The above process does not require manual intervention and can automatically adjust the baking temperature, so that the application can realize high-precision and automatic baking temperature adjustment. The application also provides a baking temperature adjusting system of a coating production line, a storage medium and an electronic device, which have the above beneficial effects, and details are not repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the application, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0065] Figure 1 A flowchart of a baking temperature adjusting method of a coating production line provided by an embodiment of the application;

[0066] Figure 2 A structural schematic diagram of an energy consumption intelligent adjusting system of a coating production line provided by an embodiment of the application;

[0067] Figure 3 A structural schematic diagram of a two-dimensional visual detection module provided by an embodiment of the application;

[0068] Figure 4 A structural schematic diagram of a three-dimensional visual point cloud processing module provided by an embodiment of the application;

[0069] Figure 5 A camera and lighting device arrangement manner schematic diagram of an image acquisition and processing module provided by an embodiment of the application;

[0070] Figure 6 A structural schematic diagram of a YOLOv7-CSM model provided by an embodiment of the application;

[0071] Figure 7 A structural schematic diagram of a contrast learning module provided by an embodiment of the application;

[0072] Figure 8 A working scene schematic diagram of a three-dimensional laser camera module provided by an embodiment of the application;

[0073] Figure 9 A numbering alignment method provided by an embodiment of the present application is shown in the figure;

[0074] Figure 10 A structure diagram of a baking energy consumption intelligent adjustment module provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0075] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0076] Please see the following Figure 1 , Figure 1 A flowchart of a baking temperature adjustment method of a coating production line provided by an embodiment of the present application is shown in the figure.

[0077] The specific steps can include:

[0078] S101: Obtain a two-dimensional image of a workpiece hanging area of a coating production line, and generate a workpiece category detection result of each workpiece in the two-dimensional image by using a target detection algorithm model.

[0079] The embodiment can be applied to a control end of a coating production line. The coating production line can include workstations such as loading, pretreatment, spraying, baking, cooling, and unloading, and the workpiece hanging area is a workstation for baking in the coating production line. The coating production line can be a coating production line of an engineering device, and the engineering device can include excavators, cranes, and shield machines.

[0080] The embodiment can set a two-dimensional camera on the coating production line, and use the two-dimensional camera to shoot a two-dimensional image of the workpiece hanging area. After obtaining the two-dimensional image, the two-dimensional image can be input into the target detection algorithm model, and then a workpiece category detection result of each workpiece in the two-dimensional image is generated. The workpiece category detection result is the type of the workpiece contained in the two-dimensional image generated by the target detection algorithm model, such as a boom, a stick, a cylinder, and a connecting rod.

[0081] The target detection algorithm model is an artificial intelligence model capable of positioning and classifying workpieces in an image. There can also be an operation of training the artificial intelligence model before this step.

[0082] S102: Obtain a three-dimensional point cloud of the workpiece hanging area, and generate a corresponding size type detection result for each workpiece in the three-dimensional point cloud.

[0083] The painting production line can also be provided with a three-dimensional laser camera to obtain a three-dimensional point cloud of the workpiece hanging area. In order to avoid missing detection of the workpiece, the photographing direction of the three-dimensional laser camera can be different from the photographing direction of the two-dimensional camera. After obtaining the three-dimensional point cloud, the workpiece size detection can be performed on the three-dimensional point cloud, and then the size type detection result of each workpiece is obtained.

[0084] The size type detection result is used to describe the type of workpiece size, such as large-size workpiece, medium-size workpiece, small-size workpiece, etc. Other rules can also be used to divide the type of workpiece size in the embodiment.

[0085] S103: Determine the total quality of the batch of workpieces according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area.

[0086] The embodiment can pre-store the corresponding relationship among the workpiece category, the workpiece size, and the workpiece quality in the database. On this basis, the embodiment can determine the quality of a single workpiece according to the workpiece category detection result, determine the quality of a single workpiece according to the size type detection result, or determine the quality of a single workpiece according to the workpiece category detection result and the size type detection result. On the basis of obtaining the quality of all workpieces contained in the two-dimensional image and the three-dimensional point cloud, the embodiment can add the quality of all workpieces to obtain the total quality of the batch of workpieces.

[0087] Specifically, if a workpiece exists in the two-dimensional image but does not exist in the three-dimensional point cloud, it is determined that the workpiece category detection result corresponding to the workpiece exists, but the size type detection result corresponding to the workpiece does not exist. At this time, the quality of the workpiece can be determined according to the workpiece category detection result.

[0088] If a workpiece exists in the three-dimensional point cloud but does not exist in the two-dimensional image, it is determined that the size type detection result corresponding to the workpiece exists, but the workpiece category detection result corresponding to the workpiece does not exist. At this time, the quality of the workpiece can be determined according to the size type detection result.

[0089] If a workpiece exists in both the two-dimensional image and the three-dimensional point cloud, it is determined that the workpiece category detection result and the size type detection result corresponding to the workpiece exist. At this time, it can be further determined whether the workpiece category detection result and the size type detection result of the workpiece match (i.e., whether they conform to the corresponding relationship among the workpiece category, the workpiece size, and the workpiece quality stored in the database). If they match, the quality of the workpiece is determined according to the workpiece category detection result and the size type detection result. If they do not match, the quality of the workpiece is determined according to the information with greater confidence in the workpiece category detection result and the size type detection result.

[0090] S104: Determine a target baking temperature corresponding to the total mass of the current batch of workpieces according to the total mass of the current batch of workpieces, and control the coating production line to perform a corresponding baking operation according to the target baking temperature.

[0091] In the embodiment, the corresponding relationship between the total mass of the workpieces and the baking temperature can be pre-stored, and the target baking temperature corresponding to the total mass of the current batch of workpieces can be determined based on the corresponding relationship. In addition, the temperature prediction model can be pre-trained, the temperature prediction model learns the mapping relationship between the total mass of the workpieces and the optimal baking temperature, and the target baking temperature corresponding to the total mass of the current batch of workpieces can be obtained by inputting the total mass of the current batch of workpieces into the temperature prediction model. After obtaining the target baking temperature, the hot air machine can be controlled to perform the baking operation on the workpieces in the coating production line according to the target baking temperature.

[0092] In the embodiment, the two-dimensional image and the three-dimensional point cloud of the workpiece hanging area are obtained, the category of the workpiece in the workpiece hanging area is determined according to the two-dimensional image, and the size of the workpiece in the workpiece hanging area is determined according to the three-dimensional point cloud. The mass of the workpiece is related to the category and size of the workpiece. Based on the category detection result and the size type detection result of the workpiece, the total mass of the current batch of workpieces is determined, and then the target baking temperature corresponding to the total mass of the current batch of workpieces is determined, so that the corresponding baking operation is performed according to the target baking temperature. The above process does not require manual intervention and can automatically adjust the baking temperature. Therefore, the embodiment can realize high-precision and automatic baking temperature adjustment.

[0093] As for Figure 1 Further to the corresponding embodiment, the coating production line is provided with a workpiece hanging rack, a three-dimensional laser camera and a plurality of two-dimensional cameras. The workpiece hanging area is an area below the workpiece hanging rack. All the two-dimensional cameras are in the same plane, and the shooting direction of the three-dimensional laser camera is perpendicular to the shooting direction of each two-dimensional camera.

[0094] The shape of the workpiece hanging rack can be rectangular. The embodiment can be preset at a distance below the workpiece hanging rack a virtual rectangle identical in shape to the workpiece hanging rack, and the virtual rectangle is parallel to the plane in which the workpiece hanging rack is located. The two-dimensional cameras used in the embodiment can include an upper left camera, a lower left camera, an upper right camera, a lower right camera, an upper edge side camera and a lower edge side camera. The camera optical centers of the upper left camera, the lower left camera, the upper right camera, the lower right camera, the upper edge side camera and the lower edge side camera are all in the plane in which the virtual rectangle is located, and the shooting directions of the upper left camera, the lower left camera, the upper right camera, the lower right camera, the upper edge side camera and the lower edge side camera all pass through the center of the diagonal line of the virtual rectangle.

[0095] The three-dimensional laser camera can be arranged on a movable device below the workpiece hanging rack, and the shooting direction of the three-dimensional laser camera is perpendicular to the plane in which the workpiece hanging rack is located.

[0096] On the basis of the above structure, the embodiment can obtain a two-dimensional image of the workpiece hanging area of the coating production line by the following manner: using all the two-dimensional cameras to shoot partial images of the workpiece hanging area of the coating production line from multiple directions, and splicing the partial images to obtain the two-dimensional image.

[0097] On the basis of the above structure, the embodiment can obtain a three-dimensional point cloud of the workpiece hanging area by the following manner: controlling the three-dimensional laser camera to move to a preset position in sequence, and obtaining a partial point cloud of the workpiece hanging area at the preset position; and splicing the partial point cloud to obtain a three-dimensional point cloud of the workpiece hanging area.

[0098] The above manner can realize multi-view imaging, thereby avoiding the problem of mutual occlusion of workpieces due to the difference in viewing angle.

[0099] As for the above Figure 1 For further introduction of the corresponding embodiment, on the basis of obtaining a three-dimensional point cloud, the embodiment can generate a corresponding size type detection result for each workpiece in the three-dimensional point cloud by the following manner: performing point cloud filtering and segmentation operation on the three-dimensional point cloud to obtain a workpiece point cloud of each workpiece; calculating a point cloud plane equation of each workpiece according to the workpiece point cloud, and determining an inclination angle of the point cloud plane equation; determining a minimum circumscribed box of the workpiece point cloud according to the point cloud plane equation, and calculating a volume of the workpiece point cloud according to the minimum circumscribed box; and determining a size type detection result of each workpiece according to the area of the workpiece point cloud, the volume of the workpiece point cloud, and the inclination angle of the point cloud plane equation.

[0100] In the three-dimensional point cloud, there can be any number of planes for a single workpiece. The embodiment can generate a corresponding point cloud plane equation for each plane of the workpiece, and then determine a minimum circumscribed box of the workpiece point cloud in combination with all the point cloud plane equations corresponding to the workpiece. The above minimum circumscribed box is the smallest circumscribed cube that can enclose the workpiece point cloud. The size estimate value of each workpiece is determined based on the area of the workpiece point cloud, the volume of the workpiece point cloud, and the inclination angle of the point cloud plane equation, and then the corresponding size type detection result is obtained.

[0101] Specifically, the embodiment can obtain a plurality of position three-dimensional point clouds, and the process of point cloud filtering and segmentation operation on the three-dimensional point clouds is as follows: performing point cloud registration splicing preprocessing operation on the three-dimensional point clouds collected at a plurality of positions to obtain integrated point clouds of a workpiece hanging area; performing point cloud filtering on the integrated point clouds to filter out noise point clouds and retain point clouds of a trolley plane and the workpiece; taking a three-dimensional camera coordinate system as a reference system, setting a Z-axis threshold, partitioning the filtered point clouds in a Z-axis direction according to the Z-axis threshold, obtaining point clouds near the trolley plane, fitting the point clouds near the trolley plane to obtain a plane equation of the trolley plane by using a plane fitting algorithm. Set a distance threshold, calculate the distance from the integrated point cloud to the trolley plane equation, and screen the point cloud coordinate index within the distance threshold range; use the screened trolley plane point cloud coordinate index to filter the point clouds of the trolley and the hook on the integrated point cloud to obtain the point clouds of the workpiece; filtering the point clouds of the workpiece, filtering out point cloud noise, and using a point cloud clustering algorithm to segment the point clouds of the workpiece to obtain independent point clouds of each workpiece. The Z-axis direction is perpendicular to the plane where the workpiece hanging frame is located, and the trolley is a device for conveying workpieces in a coating production line.

[0102] As for Figure 1 Further to the corresponding embodiment, the embodiment can determine the total mass of the workpieces in the batch in the following manner.

[0103] Step A1: selecting a target workpiece from the workpiece hanging area.

[0104] Step A2: querying a size type corresponding to the workpiece category detection result of the target workpiece in the database to obtain a size type query result.

[0105] The database stores the size type corresponding to each workpiece category, and the size type queried from the database is taken as the size type query result in this step.

[0106] Step A3: determining whether the size type detection result of the target workpiece is the same as the size type query result; if yes, proceeding to step A4; if no, proceeding to step A5.

[0107] Step A4: determining the workpiece mass of the target workpiece according to the workpiece category detection result and the size type detection result.

[0108] Step A5: determining the two-dimensional detection confidence and the three-dimensional detection confidence of the target workpiece.

[0109] The two-dimensional detection confidence is the confidence of the workpiece category detection result, and the three-dimensional detection confidence is the confidence of the size type detection result.

[0110] Step A6: If the two-dimensional detection confidence is greater than the three-dimensional detection confidence, then determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result.

[0111] Step A7: If the two-dimensional detection confidence is less than the three-dimensional detection confidence, then determining the workpiece quality of the target workpiece according to the size type detection result.

[0112] Step A8: If the two-dimensional detection confidence is equal to the three-dimensional detection confidence, then determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type detection result.

[0113] Specifically, if the two-dimensional detection confidence is equal to the three-dimensional detection confidence, then determining a first alternative quality of the target workpiece according to the workpiece category detection result and the size type query result, determining a second alternative quality of the target workpiece according to the size type detection result, and setting an average of the first alternative quality and the second alternative quality as the workpiece quality of the target workpiece.

[0114] Step A9: Determining whether all workpieces in the workpiece hanging area have been selected; if yes, then calculating the total quality of the batch of workpieces according to the workpiece qualities of all workpieces in the workpiece hanging area; if no, then entering Step A1 of selecting a target workpiece from the workpiece hanging area.

[0115] As a feasible implementation, after selecting a target workpiece from the workpiece hanging area, if the target workpiece has a corresponding workpiece category detection result and does not have a corresponding size type detection result, then determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result; if the target workpiece has a corresponding size type detection result and does not have a corresponding workpiece category detection result, then determining the workpiece quality of the target workpiece according to the size type detection result.

[0116] As a feasible implementation, the process of determining the workpiece quality of the target workpiece according to only the size type detection result includes: querying alternative workpieces matching the size type detection result in a database; and setting an average quality of all the alternative workpieces as the workpiece quality of the target workpiece.

[0117] As for the workpiece quality of the target workpiece, the workpiece quality of the target workpiece can be determined according to the workpiece category detection result and the size type query result, or can be determined according to the size type detection result. Figure 1Further to the above embodiments, before generating the workpiece category detection result of each workpiece in the two-dimensional image by using the target detection algorithm model, a training sample can also be obtained from the data set; and the target detection algorithm model is trained by using the training sample. Each training sample includes an original sample, a positive sample and a negative sample. The original sample is an image containing a workpiece, the positive sample is an image within a bounding box of the workpiece in the original sample, and the negative sample is an image outside the bounding box of the workpiece in the original sample.

[0118] The target detection algorithm model includes a backbone network, a neck network, a detection head, a contrast learning module and a loss calculation module. The contrast learning module includes a global average pooling layer, a multi-layer perceptron and a spatial attention layer. The contrast learning module is configured to convert a feature map input by the backbone network into a positive sample global enhanced feature map.

[0119] The process in which the contrast learning module converts the feature map input by the backbone network into the positive sample global enhanced feature map includes: receiving an original sample feature map F, a positive sample feature map F+ and a negative sample feature map F- output by the backbone network. The original sample feature map F, the positive sample feature map F+ and the negative sample feature map F- are input into the global average pooling layer GAP to obtain global channel weights; wherein the global channel weights include an original sample global channel weight , a positive sample global channel weight and a negative sample global channel weight . The original sample feature map F is point multiplied with the original sample global channel weight to obtain a global enhanced feature map . The original sample global channel weight , the positive sample global channel weight and the negative sample global channel weight are input into the multi-layer perceptron to obtain a processing result. The original sample feature map F, the positive sample feature map F+ and the negative sample feature map F- are input into the spatial attention layer SA to obtain spatial local feature weights; wherein the spatial local feature weights include a positive sample spatial local feature weight , a negative sample spatial local feature weight and an original sample spatial local feature weight . The original sample feature map F, the positive sample feature map F+ and the negative sample feature map F- are respectively point multiplied with corresponding spatial local feature weights to obtain local enhanced feature maps; wherein the local enhanced feature maps include an original sample local enhanced feature map Fs, a positive sample local enhanced feature map and a negative sample local enhanced feature map The original sample local enhanced feature map Fs is subjected to element subtraction operation with the negative sample local enhanced feature map An element subtraction operation is performed on the local information attention feature Fsl to obtain a local information attention feature Fsl. A contrast loss function value is calculated according to the processing result and the local enhanced feature map, and a similarity loss function value is calculated according to the positive sample local enhanced feature map and the local information attention feature, so as to optimize the parameters of the contrast learning module by using the contrast loss function value and the similarity loss function value. An element addition operation is performed on the global enhanced feature map and the local information attention feature Fsl to obtain a positive sample global enhanced feature map ; wherein the positive sample global enhanced feature map is used as input to the neck network.

[0120] The skeleton network can output sample feature maps of multiple spatial resolutions for three types of samples (original samples, positive samples and negative samples), and take sample feature maps of the same spatial resolution as a group of original sample feature maps F, positive sample feature maps F+ and negative sample feature maps F- so that the contrast learning module processes original sample feature maps F, positive sample feature maps F+ and negative sample feature maps F- of the same spatial resolution.

[0121] As for Figure 1 Further to the corresponding embodiments, before determining the corresponding target baking temperature according to the total mass of the workpieces in the current batch, a mapping relationship between the total mass of the workpieces and the baking temperature can be determined according to historical data, a temperature prediction model can be established according to the mapping relationship, and the temperature prediction model can be optimized by using a gradient descent algorithm.

[0122] Correspondingly, determining the corresponding target baking temperature according to the total mass of the workpieces in the current batch and controlling the coating production line to perform a corresponding baking operation according to the target baking temperature includes: inputting the total mass of the workpieces in the current batch into the temperature prediction model to obtain a target baking temperature; and adjusting the opening degree of the natural gas regulating valve according to the target baking temperature so as to control the hot air fan to perform a baking operation on the workpieces in the coating production line according to the target baking temperature.

[0123] In the process of baking the workpiece, the existing scheme usually determines the baking temperature according to the working parameters of the hot air machine, but in actual application, due to the change of environmental humidity, the difference of initial temperature of the workpiece, the power fluctuation caused by aging of the hot air machine and the uneven thickness of the coating on the surface of the workpiece, it is difficult to ensure that the workpiece is actually baked at the target baking temperature; In view of this problem, the embodiment has the improvement mode in this aspect, as follows: an infrared thermal imager, a temperature sensor and a humidity sensor are arranged on the coating production line, the infrared thermal imager is used to detect the surface temperature of the workpiece, the temperature sensor is used to detect the environmental temperature, and the humidity sensor is used to detect the environmental humidity; The embodiment can monitor the surface temperature of the workpiece, the environmental temperature and the environmental humidity in real time during the baking process of the coating production line, and process the surface temperature of the workpiece, the environmental temperature and the environmental humidity by using Kalman filtering or fuzzy fusion model to generate baking state feedback signal, so as to dynamically adjust the target baking temperature according to the baking state feedback signal through the adaptive algorithm.

[0124] The above-mentioned process described in the embodiment is illustrated by the following examples in practical application.

[0125] With the rapid development of artificial intelligence and intelligent manufacturing technology, the factory production mode of the engineering machinery industry is transforming towards automation and unmanned direction, and the industry also puts forward higher requirements for the application of intelligent manufacturing technology in the green, energy-saving, intelligent and automatic aspects of the industrial field.

[0126] Excavators are the mainstream products in the engineering machinery industry, and their assembly production lines are one of the core links of factory manufacturing. However, in the existing production mode, the baking temperature adjustment for each batch of excavator workpieces in the traditional excavator workpiece coating and baking operation process still relies on manual counting and experience estimation setting. Although this method has low labor cost, it has low production efficiency, insufficient automation, and the manual temperature determination cannot be standardized, resulting in energy waste and increased factory emissions.

[0127] The key to solving this problem lies in accurately identifying the category and size of each workpiece and automatically setting the accurate baking temperature. Although the target detection algorithm based on deep learning (such as YOLO series) is widely used in the industry. However, in the scene of excavator workpiece coating and baking, its recognition accuracy and generalization ability are still challenged, specifically: when the workpiece passes through the crane suspension into the baking area, the workpiece size is different and the distance is close, the field of view of a single two-dimensional camera is limited, resulting in occlusion of the workpiece image and difficulty in obtaining complete information. In addition, excavator workpieces have the characteristics of cross-scale, various types and similar features, and YOLO (You Only Look Once, a target detection algorithm) based on two-dimensional monocular camera cannot adapt to these complex situations.

[0128] In the prior art, a three-dimensional reconstruction technology is used to overcome the problem of field of view loss through binocular or multi-view cameras, but this type of technology requires expensive industrial-grade CCD (Charge Coupled Device) cameras and high-precision multi-target positioning algorithms. In actual production, the area of the workpiece suspended by the vehicle is large, and the space available for the camera is limited, resulting in a large difference in field of view scale between cameras that are far apart, and there is a non-overlapping field of view, which restricts the accuracy of the multi-target positioning algorithm. There is also an object recognition technology based on three-dimensional vision, which directly scans the workpiece area through a three-dimensional imaging method, and uses a point cloud processing algorithm to recognize the type and geometric features of the workpiece, but this type of technology still has the following problems: traditional point cloud processing schemes based on PCL (Point Cloud Library) rely on template registration technology to determine the type of workpiece, but in reality, it is difficult to fully cover the 3D (3 Dimensional) model library of the workpiece, and the production cycle is long and the cost is high. Point cloud recognition algorithms based on deep learning also face the dilemma of high complexity in training data set production. In addition, point cloud volume estimation algorithms based on sectional area integration are not effective in the case of hollow and irregular structures inside the workpiece.

[0129] It can be seen that the target detection algorithm based on monocular vision is difficult to cope with the situation of large detection scene span, occlusion between workpieces, multi-scale, diversity, etc. The three-dimensional reconstruction technology based on binocular or multi-view cameras is expensive, and it is difficult to overcome the problem of insufficient reconstruction accuracy caused by non-overlapping field of view. The object recognition technology based on three-dimensional vision is limited by the challenge of varying geometric features of the workpiece, and the difficulty of data or 3D model production.

[0130] To solve the above problems, the embodiment provides an energy consumption intelligent adjustment scheme for an excavator coating production line, solves the accurate and general identification problem of the size type and category of the excavator workpiece in the excavator workpiece coating and baking scene, and overcomes the low automation and energy consumption waste problem of the traditional manual setting of the baking temperature. The embodiment can integrate the respective advantages of two-dimensional and three-dimensional vision, overcome the problems of limited field of view and insufficient model robustness of monocular target detection algorithm, realize effective multi-view perception and detection of the workpiece in an economical way, design a simple and effective three-dimensional vision point cloud processing algorithm to solve the misjudgment problem caused by the same features between workpieces under two-dimensional imaging and lack of three-dimensional information, realize accurate evaluation of the workpiece category in the excavator coating and baking scene by combining two-dimensional and three-dimensional information, and automatically and accurately adjust the baking temperature of the batch workpiece through the baking temperature intelligent adjustment algorithm. The embodiment designs a network model YOLOv7-CSM based on YOLOv7 and contrastive learning, and improves the ability of the model to distinguish between workpieces and backgrounds in the case of workpiece overlap by constructing a CSM module. The embodiment proposes a multi-view imaging scheme composed of multiple two-dimensional cameras to overcome the problem of mutual occlusion of workpieces caused by different viewing angles. The high cost and limited calibration accuracy problems existing in the three-dimensional reconstruction scheme based on multiple cameras are avoided. The embodiment designs a point cloud processing algorithm based on three-dimensional vision, and does not introduce a template registration and deep learning model, and obtains the basic size type of the excavator workpiece by evaluating the geometric shape, area and volume of the workpiece placement. The embodiment proposes a scheme for matching the target detection type of the workpiece under the imaging of the multi-view two-dimensional camera and the size type of the workpiece point cloud under the three-dimensional vision and mixed decision of the workpiece category, integrates and refines the evaluation of the two-dimensional detection result and the three-dimensional detection result, and outputs the accurate workpiece category. The embodiment proposes an intelligent baking energy consumption adjustment scheme to realize intelligent and automatic adjustment of the baking temperature of the batch workpiece of the excavator coating production line and reduce energy consumption waste.

[0131] Referring to Figure 2 , Figure 2 is a structural schematic diagram of an energy consumption intelligent adjustment system of a coating production line provided by the embodiment of the present application. The system comprises a two-dimensional vision detection module, a three-dimensional vision point cloud processing module, and a baking energy consumption intelligent adjustment module. A control end (such as an industrial computer or a server) of the coating production line is connected with the two-dimensional vision detection module and the three-dimensional vision point cloud processing module respectively.

[0132] Referring to Figure 3 , Figure 3A structure diagram of a two-dimensional visual detection module provided by an embodiment of the present application, the two-dimensional visual detection module comprising an image acquisition and processing module and a target detection algorithm model, the image acquisition and processing module comprising a lighting device, a plurality of two-dimensional cameras (such as camera 1, camera 2, …, camera 6) and an image preprocessing module, the image acquisition and processing module can acquire a two-dimensional image of a workpiece of a mining machine, and transmit the two-dimensional image to the target detection algorithm model. An industrial computer or a server is connected with the image acquisition and processing module and the target detection algorithm model respectively.

[0133] Please refer to Figure 4 , Figure 4 A structure diagram of a three-dimensional visual point cloud processing module provided by an embodiment of the present application, the three-dimensional visual point cloud processing module comprising a three-dimensional laser camera module, a point cloud processing algorithm module and a hybrid decision module, the three-dimensional laser camera module comprising a lighting device, an electric cart and a three-dimensional laser camera. The three-dimensional laser camera module is used for acquiring a three-dimensional point cloud of a workpiece of a mining machine, and an industrial computer or a server is connected with the two-dimensional visual detection module, the hybrid decision module, the point cloud processing algorithm module and the three-dimensional laser camera module respectively.

[0134] The image acquisition and processing module is mainly used for constructing multi-view two-dimensional imaging of a workpiece of a mining machine, acquiring image information of the workpiece under different fields of view, and preliminarily processing the acquired data to provide data support for a subsequent target detection algorithm model. In the embodiment, the hardware of the image acquisition and processing module mainly comprises six cameras and a lighting device, wherein the cameras are respectively located at the upper left, the lower left, the upper right, the lower right, the upper side and the lower side of the workpiece to be detected.

[0135] Please refer to Figure 5 , Figure 5 A camera and lighting device arrangement manner diagram of an image acquisition and processing module provided by an embodiment of the present application, the two-dimensional camera used in the embodiment comprising an upper left camera D1, a lower left camera D2, an upper right camera D3, a lower right camera D4, an upper side camera D5 and a lower side camera D6; the upper left camera D1, the lower left camera D2, the upper right camera D3 and the lower right camera D4 have corresponding lighting devices, namely an upper left light source B1, a lower left light source B2, an upper right light source B3 and a lower right light source B4. Figure 5 In the figure, C1 represents a workpiece suspension frame (i.e. a workpiece suspension row frame), and C2 represents a same-length-and-width rectangle below the workpiece suspension frame.

[0136] The left upper camera D1, the left lower camera D2, the right upper camera D3 and the right lower camera D4 are respectively located at the vertical downward 500mm discharge port and the vertical downward 500mm inlet port of the workpiece hanging rack C1, and the camera optical center is directed to the diagonal center of the vertical downward 500mm same length and width rectangle, the upper side camera D5 and the lower side camera D6 are respectively located at the vertical downward 500mm same length and width rectangle upper side 500mm and the vertical downward 500mm same length and width rectangle lower side 500mm, and the camera optical center is directed to the diagonal center of the rectangle, and the lighting device includes four light sources, wherein each light source is respectively located at the vertical upward 500mm of the left upper camera D1, the left lower camera D2, the right upper camera D3 and the right lower camera D4, and the light source optical center is directed to the diagonal center of the vertical downward 500mm same length and width rectangle at 45 degrees downward.

[0137] The data preprocessing method steps of the image acquisition and processing module in the embodiment are as follows:

[0138] Step E1: The two-dimensional images collected in the left upper, left lower, right upper and right lower four directions are divided into a pair of left upper and left lower and a pair of right upper and right lower.

[0139] Step E2: The key point features of the images are calculated by Harris (Harris operator) for each pair of images.

[0140] Step E3: The invariant feature descriptors of the two images are calculated by SIFT (Scale-Invariant Feature Transform), and the two images are registered according to the key point features and the invariant feature descriptors.

[0141] Step E4: The homography matrix between the features of the two images is calculated by Ransac (Random Sample Consensus).

[0142] Step E5: The images are subjected to affine transformation by the homography matrix, and finally the left workpiece image and the right workpiece image are obtained by splicing and fusing the current pair of images.

[0143] The target detection algorithm model is used to calculate the two-dimensional images (i.e., the left workpiece image and the right workpiece image) obtained by the image acquisition and processing module, and obtain the category of the workpiece. In the embodiment, the target detection algorithm model is improved based on the YOLOv7 (a kind of target detection algorithm) model, and a contrast learning module CSM is designed to enhance the model's ability to distinguish targets and backgrounds and improve the model's precision and generalization ability.

[0144] The embodiment can label the collected two-dimensional image, then insert the designed contrast learning module on the YOLOv7 model to obtain a YOLOv7-CSM (i.e., a target detection algorithm provided with a contrast learning module) model, train the YOLOv7-CSM model, and finally perform inference on the YOLOv7-CSM model to obtain the category of the workpiece in the image.

[0145] The embodiment can divide the image data set collected by the image acquisition and processing module into a training set, a validation set, and a test set according to a ratio of 3:1:1. The embodiment can also label the image data; specifically, the data in the data set can be divided into three types of large, medium, and small according to the specific size of the excavator workpiece, and the independent workpieces in the three types of large, medium, and small are labeled in detail, and the labeling information includes: workpiece category, workpiece bounding box corner coordinates, and workpiece size type. According to the coordinates of the bounding box of the labeling target, the positive and negative sample images are divided; specifically, for the positive sample, only the pixel values within the target bounding box are retained, and other pixel values outside the box are assigned a value of 0. For the negative sample, only the pixel values outside the target bounding box are retained, and other pixel values outside the box are assigned a value of 0. The original image (i.e., the original sample), the positive sample, and the negative sample are paired to obtain the excavator workpiece data set for training. The embodiment can perform image enhancement operations on the data set before feeding it into the model for training, including: image scaling, image grayscale processing, contrast enhancement, etc.

[0146] The embodiment can divide the image data set collected by the image acquisition and processing module into a training set, a validation set, and a test set according to a ratio of 3:1:1. The embodiment can also label the image data; specifically, the data in the data set can be divided into three types of large, medium, and small according to the specific size of the excavator workpiece, and the independent workpieces in the three types of large, medium, and small are labeled in detail, and the labeling information includes: workpiece category, workpiece bounding box corner coordinates, and workpiece size type. According to the coordinates of the bounding box of the labeling target, the positive and negative sample images are divided; specifically, for the positive sample, only the pixel values within the target bounding box are retained, and other pixel values outside the box are assigned a value of 0. For the negative sample, only the pixel values outside the target bounding box are retained, and other pixel values outside the box are assigned a value of 0. The original image (i.e., the original sample), the positive sample, and the negative sample are paired to obtain the excavator workpiece data set for training. The embodiment can perform image enhancement operations on the data set before feeding it into the model for training, including: image scaling, image grayscale processing, contrast enhancement, etc. Figure 6 The embodiment can construct an improved YOLOv7 network structure: YOLOv7-CSM.

[0147] Figure 6 A structure diagram of a YOLOv7-CSM model provided by the embodiment of the present application, the YOLOv7-CSM model includes a contrast learning module, the contrast learning module is located after feature maps P3, P4, and P5 output by a skeleton network of YOLOv7, and the contrast learning module is located before a neck network Neck structure. The above-mentioned model also includes a backbone network Backbone, a detection head Head, and a loss calculation module Loss, and the input of the backbone network includes an original sample Img, a positive sample Img+, and a negative sample Img-. After the original sample Img is processed by the backbone network, an original sample feature map F is obtained; after the positive sample Img+ is processed by the backbone network, a positive sample feature map F+ is obtained; and after the negative sample Img- is processed by the backbone network, a negative sample feature map F- is obtained. The feature maps P3, P4, and P5 are three feature maps with different spatial resolutions.

[0148] Specifically, the contrast learning module includes a global average pooling layer GAP, a multilayer perceptron MLP, a spatial attention layer SA, two point multiplication and element addition and subtraction operations, a contrast loss function, and an approximation loss function.

[0149] The contrast loss function used by the contrast learning module The formula is as follows.

[0150] .

[0151] wherein, is the approximation loss function, and here a cosine similarity loss is used, N represents the total number of a set of positive sample feature maps, negative sample feature maps, and original sample feature maps, is a temperature parameter for function adjustment.

[0152] represents a contrast loss function value, represents an expectation, represents a logarithm, k and i represent sample serial numbers, represents an indicator function, represents the kth original sample feature map.

[0153] The approximation loss function used The formula is as follows.

[0154] .

[0155] wherein is a feature flattening operation, which reduces the dimensionality of the feature map, and n represents the total length of the feature after dimensionality reduction.

[0156] represents an approximation loss function value.

[0157] See Figure 7 , Figure 7 is a structure diagram of a contrast learning module provided by an embodiment of the present application, wherein represents a point multiplication operation, represents an element subtraction, represents an element addition, and the specific steps of the contrast learning module for processing the feature map are as follows:

[0158] Step F1: for each feature map (feature map P3, feature map P4, and feature map P5) of the original sample, take the feature maps at the corresponding positions of the positive sample and the negative sample in the same batch to form a set of original sample feature maps F, positive sample feature maps F+, and negative sample feature maps F-, and send them into the contrast learning module.

[0159] Step F2: The original sample feature map F, the positive sample feature map F+, and the negative sample feature map F- respectively enter the global average pooling layer GAP and the spatial attention layer SA, to extract global channel weights (including original sample global channel weights , positive sample global channel weights , and negative sample global channel weights ) and spatial local feature weights (including positive sample spatial local feature weights , negative sample spatial local feature weights , and original sample spatial local feature weights ).

[0160] Wherein, after the original sample feature map F, the positive sample feature map F+, and the negative sample feature map F- are input into the global average pooling layer GAP, the global average pooling layer GAP outputs the original sample global channel weights corresponding to the original sample feature map, the positive sample global channel weights corresponding to the positive sample feature map, and the negative sample global channel weights corresponding to the negative sample feature map.

[0161] After the original sample feature map F, the positive sample feature map F+, and the negative sample feature map F- are input into the spatial attention layer SA, the spatial attention layer SA outputs the positive sample spatial local feature weights corresponding to the positive sample feature map, the negative sample spatial local feature weights corresponding to the negative sample feature map, and the original sample spatial local feature weights corresponding to the original sample feature map.

[0162] Step F3: After the original sample feature map F and the original sample global channel weights are multiplied, the global enhanced feature map is obtained, and the original sample global channel weights , the positive sample global channel weights , and the negative sample global channel weights obtained in step F2 are input into the multi-layer perceptron MLP and then into the contrastive loss function for calculation.

[0163] Step F4: The original sample feature map F, the positive sample feature map F+, and the negative sample feature map F- are respectively multiplied by the original sample spatial local feature weights , the positive sample spatial local feature weights , and the negative sample spatial local feature weights to obtain the original sample local enhanced feature map Fs, the positive sample local enhanced feature map , and the negative sample local enhanced feature map Fs, the positive sample local enhanced feature map Fs, and the negative sample local enhanced feature map into the contrast loss function for calculation.

[0164] Specifically, the original sample feature map F and the original sample spatial local feature weight are subjected to dot multiplication operation to obtain the original sample local enhanced feature map Fs; the positive sample feature map F+ and the positive sample spatial local feature weight are subjected to dot multiplication operation to obtain the positive sample local enhanced feature map ; the negative sample feature map F- and the negative sample spatial local feature weight are subjected to dot multiplication operation to obtain the negative sample local enhanced feature map .

[0165] Step F5: taking the original sample local enhanced feature map Fs, the positive sample local enhanced feature map and the negative sample local enhanced feature map obtained in step F4, subtracting the negative sample local enhanced feature map from the original sample local enhanced feature map Fs to obtain the local information focus feature Fs1, and taking the positive sample local enhanced feature map and Fs1 into the approximation loss function for calculation.

[0166] Step F6: taking obtained in step F3 and Fs1 obtained in step F5 to perform element addition operation to obtain the positive sample global enhanced feature map , and taking the positive sample global enhanced feature map into the neck network.

[0167] The YOLOv7-CSM model is trained, and the performance of the trained model is evaluated using the test set. Based on the trained YOLOv7-CSM model, model inference is performed to obtain the workpiece detection class result in the image.

[0168] In the present application, the three-dimensional laser camera module is mainly used to obtain three-dimensional point cloud data of the workpiece, and the module includes a lighting device, an electric cart and a three-dimensional laser camera, wherein the lighting device and the three-dimensional laser camera are located on the cart.

[0169] Please refer to Figure 8 , Figure 8A working scene schematic diagram of a three-dimensional laser camera module provided by an embodiment of the present application is shown in the figure. G1 represents a three-dimensional laser camera, G2 represents a lighting device, G3 represents an electric trolley, H1 represents a left area, H2 represents a middle area, and H3 represents a right area. The figure also shows an entrance side, an exit side, and a plurality of workpieces.

[0170] The embodiment can divide the workpiece hanging area into three collection areas, i.e., a right area, a middle area, and a left area, according to the right, middle, and left three hole rectangles of the workpiece hanging rack, and take the intersection of the diagonal lines of the rectangles as the center. The three-dimensional laser camera is turned on, initialized, and calibrated. After the YOLOv7-CSM model detects the workpiece, the server transmits a start flag to the control processing unit on the trolley. The trolley travels at a constant speed and moves from the right area to the left area of the workpiece hanging area in turn. Each time, the trolley stops at the center of each area and transmits the point cloud of the area to the server. After the shooting of the three areas is completed, the control processing unit on the trolley transmits a point cloud processing algorithm running flag back to the algorithm module in the server. Then, the trolley with the three-dimensional camera returns to the initial position at a constant speed and waits for the next scanning task.

[0171] The point cloud processing algorithm module is mainly used for processing the workpiece point cloud data collected by the three-dimensional laser camera module, evaluating the geometric features of each workpiece point cloud, such as volume, area, and inclination, and identifying the size and type of each workpiece, such as large, medium, and small.

[0172] The processing process of the point cloud processing algorithm module includes the following steps.

[0173] Step I1: The received point cloud data collected in the three areas is subjected to point cloud registration splicing preprocessing to obtain integrated point cloud of the workpiece hanging area.

[0174] Step I2: The integrated point cloud is subjected to point cloud filtering to filter out noise point cloud and retain the point cloud of the trolley plane and the workpiece.

[0175] Step I3: Taking the three-dimensional camera coordinate system as the reference system, a Z-axis threshold is set, the filtered point cloud is partitioned in the Z-axis direction according to the Z-axis threshold, the point cloud near the trolley plane is obtained, the plane fitting algorithm RANSAC is used to fit the point cloud near the trolley plane to obtain the plane equation of the trolley, a distance threshold is set, the distance of the integrated point cloud to the trolley plane equation is calculated, and the point cloud coordinate index within the distance threshold is screened.

[0176] Step I4: Using the screened trolley plane point cloud coordinate index, the point cloud of the trolley and the hook is filtered on the integrated point cloud to obtain the point cloud of the workpiece.

[0177] Step I5: The workpiece point cloud is subjected to filtering to filter out point cloud noise, and the point cloud clustering algorithm is used to segment the workpiece point cloud to obtain the independent point cloud of each workpiece.

[0178] Step I6: For each workpiece point cloud, the point cloud plane equation is fitted, and the inclination of the point cloud plane equation of each workpiece to the horizontal reference surface is calculated, and the horizontal reference surface is perpendicular to the Z axis of the three-dimensional camera coordinate system.

[0179] Step I7: Estimate the area of each workpiece point cloud, fit the minimum circumscribed frame of each workpiece point cloud, estimate the volume of the point cloud according to the circumscribed frame, and predict the size type (such as large size, medium size or small size) of the workpiece according to the two parameters of the obtained area and volume combined with the inclination of the workpiece point cloud plane equation obtained in the sixth step, and output the prediction confidence.

[0180] In the embodiment, the hybrid decision module is mainly used for numbering the independent point cloud of each workpiece obtained by the point cloud processing algorithm, and aligning the two-dimensional workpiece bounding box obtained by the target detection algorithm in the two-dimensional vision detection module according to the number, so as to match the two-dimensional detection information of the workpiece and the three-dimensional prediction information of the three-dimensional point cloud processing algorithm, and then infer the accurate category and size type of the workpiece through the fusion of two-dimensional and three-dimensional visual decision.

[0181] Please refer to Figure 9 , Figure 9 for a numbering alignment method provided by the embodiment of the application, wherein 1, 2, 3, 4, 5, 6, 7, 8 and 9 represent the labels of the workpieces, X, Y and Z represent the three-dimensional laser camera coordinate system, the left area, the middle area and the right area of the workpiece hanging frame are shown in the figure, the positions of the workpieces in the left view, the right view, the lower side view and the upper side view are shown in the figure, and the key position relationships include left, right, far and near.

[0182] The process of determining the workpiece category and size type by the hybrid decision module is as follows.

[0183] Step J1: According to the three-dimensional camera shooting path, the workpieces are numbered in increasing order from the right area to the middle area to the left area (that is, according to the Y axis direction of the three-dimensional laser camera), and the starting numbering order is sequentially increased from left to right according to the X axis direction of the three-dimensional laser camera, as shown in Figure 9 .

[0184] Step J2: The workpiece point cloud obtained by the point cloud processing algorithm is numbered according to the method described in step J1.

[0185] Step J3: The depth estimation neural network model is used to perform depth estimation on the left workpiece image, the right workpiece image, the upper side workpiece image and the lower side workpiece image data collected by the two-dimensional camera, to obtain the distance relationship of the pixels.

[0186] Step J4: The left workpiece image, the right workpiece image, the upper side workpiece image, and the lower side workpiece image data collected by the two-dimensional camera are input into the YOLOv7-CSM model to obtain the workpiece detection object results, which are numbered. The numbering rule is as follows: taking the workpiece point cloud number as a reference, for the right workpiece image, the workpieces are numbered from left to right and from near to far. For the left workpiece image, the workpieces are numbered from right to left and from far to near. For the upper side workpiece image, the workpieces are numbered from left to right and from far to near. For the lower side workpiece image, the workpieces are numbered from right to left and from near to far. Among them, the right workpiece image is only responsible for the workpieces in the right area, and the left workpiece image is only responsible for the workpieces in the left area.

[0187] Step J5: The workpiece detection object number detected by the two-dimensional camera is integrated and aligned with the workpiece point cloud number under the three-dimensional camera. If there is a workpiece missing or missing in the YOLOv7-CSM detection data corresponding to the left workpiece image, the right workpiece image, the upper side workpiece image, or the lower side workpiece image collected by the two-dimensional camera, the detection results of other workpiece image data are observed first. If all are missing, the workpiece point cloud number is used as a reference, and only the workpiece size type of the number is retained.

[0188] Step J6: According to the workpiece number in step J5, the corresponding point cloud size type estimation and the target detection category result of the YOLOv7-CSM model are obtained in sequence. If the workpiece point cloud type estimation and the target detection have values and the estimated workpiece size types are consistent, the consistent result is taken as the final size of the workpiece, and the target detection category result is taken as the category of the workpiece.

[0189] Among them, the above-mentioned "workpiece point cloud type estimation and target detection have values and the estimated workpiece size types are consistent" means that the size type obtained by the workpiece point cloud type estimation matches the workpiece category obtained by the target detection.

[0190] Step J7: If the workpiece point cloud size type estimation and the target detection category of the workpiece in step J6 have values but the workpiece size types are inconsistent, the result with the highest prediction confidence is taken as the standard. If the point cloud size type estimation confidence is large, only the size type is output. If the target detection category confidence is large, the workpiece size type and the category are output.

[0191] Among them, the above-mentioned "workpiece point cloud size type estimation and target detection have values but the workpiece size types are inconsistent" means that the size type obtained by the workpiece point cloud type estimation does not match the workpiece category obtained by the target detection.

[0192] Step J8: If one of the workpiece point cloud size type estimation and the target detection category in step J6 has no value, then the object with the value is taken as the final output of the workpiece type. For example, only the point cloud size type is output as the workpiece size type, and if only the target detection category is output, then the target detection category is taken as the size and category result of the workpiece.

[0193] Among them, the "one of the workpiece point cloud size type estimation and the target detection category has no value" means that there is no size type obtained by the workpiece point cloud type estimation, or there is no workpiece category obtained by the target detection.

[0194] Please refer to Figure 10 , Figure 10 The structure diagram of the baking energy consumption intelligent adjustment module provided by the embodiment of the present application. The baking energy consumption intelligent adjustment module includes a temperature adjustment module, a hot air machine, a natural gas regulating valve and other components. The industrial computer / server is connected with the temperature adjustment module and the natural gas regulating valve, and the input of the temperature adjustment module includes the workpiece type of the excavator. The temperature adjustment module is realized by two parts of quality estimation and temperature adjustment algorithm (i.e. quality and temperature mapping function model). First, according to the workpiece type of each batch obtained by the two-dimensional vision detection module and the three-dimensional vision point cloud processing module, the corresponding quality value of the workpiece is taken out from the workpiece quality database, and the total quality of the workpieces in this batch is calculated; after obtaining the total quality of the workpieces in this batch, the mapping function relationship between the quality and the temperature is established based on the gradient descent and linear regression algorithm; according to the temperature calculated by the mapping function, the industrial computer adjusts the opening degree of the natural gas regulating valve to control the blowing temperature of the hot air machine, so as to realize the temperature adjustment of the baking house. Through the module, the baking energy consumption intelligent adjustment can be automatically realized, and the energy waste caused by the fixed setting of the unified baking temperature by the artificial is reduced.

[0195] The baking energy consumption intelligent adjustment module can realize the following steps.

[0196] Step K1: quality estimation.

[0197] According to the workpiece type of each batch calculated by the two-dimensional vision detection module and the three-dimensional vision point cloud processing module, the corresponding quality value of the workpiece is taken out from the workpiece quality database established by the artificial, if only the size, the average quality value of the size is taken, and the total quality of the workpieces in this batch is calculated.

[0198] Step K2: according to the labeled data set of the workpiece quality and the temperature according to the expert experience, and the total sample number of the data set is randomly divided into a training set, a verification set and a test set according to the ratio of 3:1:1.

[0199] Step K3: quality and temperature mapping function model training.

[0200] An initial model is established for the mapping relationship between the workpiece quality and the baking temperature using a linear equation, the training set and the validation set in ② are input into the model, and the optimal weight between the quality and the temperature is obtained by using the gradient descent optimization method. Finally, the performance of the model is evaluated by the test set, and the model available for inference is obtained.

[0201] Step K4: temperature estimation inference.

[0202] The baking temperature corresponding to the total quality of the batch workpieces is inferred from the mapping function model between the quality and the temperature established in step K3.

[0203] Step K5: energy consumption adjustment.

[0204] According to the temperature inferred in step K4, the opening degree of the natural gas regulating valve is adjusted by the industrial computer to control the blowing temperature of the hot air fan, so as to realize the temperature adjustment of the baking house.

[0205] The embodiment provides a component structure design of the energy consumption intelligent adjustment method for the excavator coating production line; the embodiment provides a component structure design of the two-dimensional visual detection module, and designs different functions for each module; the embodiment provides a component structure design of the three-dimensional visual point cloud processing module, and designs different functions for each module; and the embodiment provides a component structure design of the baking energy consumption intelligent adjustment module, and designs different functions for each module. The embodiment can solve the problem that the size of each batch of workpieces is different, the hanging position of the workpiece is random, and there is single-angle occlusion between the small and medium-sized workpieces in the excavator workpiece coating production line scene, the shapes of the workpieces are different, and the types of the workpieces are various, how to accurately identify the category of each workpiece through vision, thereby automatically adjusting the baking temperature of the batch workpieces, further promoting the automation and energy saving of the excavator coating production line, reducing the energy waste and emission increase caused by manually fixing and setting the baking temperature, and promoting the intelligent and green transformation of the production line.

[0206] The coating production line baking temperature adjustment system provided in the embodiment of the application can comprise:

[0207] A two-dimensional image processing module is configured to acquire a two-dimensional image of a workpiece hanging area of the coating production line, and generate a workpiece category detection result of each workpiece in the two-dimensional image by using a target detection algorithm model.

[0208] A three-dimensional point cloud processing module is configured to acquire a three-dimensional point cloud of the workpiece hanging area, and generate a corresponding size type detection result for each workpiece in the three-dimensional point cloud.

[0209] A quality calculation module is configured to determine the total quality of the batch workpieces according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area.

[0210] a temperature control module configured to determine a target baking temperature corresponding to the total mass of the batch of workpieces according to the total mass of the batch of workpieces, and control the coating production line to perform a corresponding baking operation according to the target baking temperature.

[0211] The embodiment obtains a two-dimensional image and a three-dimensional point cloud of the workpiece hanging area, determines the category of the workpiece in the workpiece hanging area according to the two-dimensional image, and determines the size of the workpiece in the workpiece hanging area according to the three-dimensional point cloud. The mass of the workpiece is related to the category and size of the workpiece. Based on the detection result of the category of the workpiece and the detection result of the size of the workpiece, the embodiment determines the total mass of the batch of workpieces, and then determines the target baking temperature corresponding to the total mass of the batch of workpieces, so as to perform a corresponding baking operation according to the target baking temperature. The above process does not require manual intervention and can automatically adjust the baking temperature. Therefore, the embodiment can realize high-precision and automatic baking temperature adjustment.

[0212] Further, the process of determining the total mass of the batch of workpieces by the mass calculation module according to the detection result of the category of each workpiece and / or the detection result of the size of each workpiece in the workpiece hanging area includes: selecting a target workpiece from the workpiece hanging area; querying a size type corresponding to the detection result of the category of the target workpiece in the database to obtain a size type query result; determining whether the detection result of the size of the target workpiece is the same as the size type query result; if yes, determining the mass of the target workpiece according to the detection result of the category of the workpiece and the detection result of the size of the workpiece; if no, determining a two-dimensional detection confidence and a three-dimensional detection confidence of the target workpiece; the two-dimensional detection confidence is the confidence of the detection result of the category of the workpiece, and the three-dimensional detection confidence is the confidence of the detection result of the size of the workpiece; if the two-dimensional detection confidence is greater than the three-dimensional detection confidence, determining the mass of the target workpiece according to the detection result of the category of the workpiece and the size type query result; if the two-dimensional detection confidence is less than the three-dimensional detection confidence, determining the mass of the target workpiece according to the detection result of the size of the workpiece; if the two-dimensional detection confidence is equal to the three-dimensional detection confidence, determining the mass of the target workpiece according to the detection result of the category of the workpiece and the detection result of the size of the workpiece; determining whether all workpieces in the workpiece hanging area have been selected; if yes, calculating the total mass of the batch of workpieces according to the mass of all workpieces in the workpiece hanging area; if no, entering the step of selecting a target workpiece from the workpiece hanging area.

[0213] Further, the quality calculation module is further configured to: if the target workpiece has a corresponding workpiece category detection result and does not have a corresponding size type detection result, determine the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result; and if the target workpiece has a corresponding size type detection result and does not have a corresponding workpiece category detection result, determine the workpiece quality of the target workpiece according to the size type detection result.

[0214] Further, the process of determining the workpiece quality of the target workpiece according to the size type detection result by the quality calculation module includes: querying alternative workpieces matching the size type detection result in a database; and setting the average quality of all the alternative workpieces as the workpiece quality of the target workpiece.

[0215] Further, the quality calculation module is further configured to: if the target workpiece has a corresponding workpiece category detection result and does not have a corresponding size type detection result, determine the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result; and if the target workpiece has a corresponding size type detection result and does not have a corresponding workpiece category detection result, determine the workpiece quality of the target workpiece according to the size type detection result.

[0216] The model training module is configured to: before generating the workpiece category detection result of each workpiece in the two-dimensional image by using the target detection algorithm model, acquire training samples from a data set; each training sample includes an original sample, a positive sample and a negative sample, the original sample is an image containing a workpiece, the positive sample is an image within a bounding box of the workpiece in the original sample, and the negative sample is an image outside the bounding box of the workpiece in the original sample; and train the target detection algorithm model by using the training samples.

[0217] The target detection algorithm model includes a skeleton network, a neck network, a detection head, a contrast learning module and a loss calculation module, the contrast learning module includes a global average pooling layer, a multilayer perceptron and a spatial attention layer, and the contrast learning module is configured to convert a feature map input by the skeleton network into a positive sample global enhanced feature map.

[0218] Further, the process that the contrast learning module converts the feature map of the skeleton network input into a positive sample global enhanced feature map comprises: receiving original sample feature maps, positive sample feature maps and negative sample feature maps output by the skeleton network; inputting the original sample feature maps, the positive sample feature maps and the negative sample feature maps into the global average pooling layer to obtain global channel weights; wherein the global channel weights comprise original sample global channel weights, positive sample global channel weights and negative sample global channel weights; performing point multiplication on the original sample feature maps and the original sample global channel weights to obtain a global enhanced feature map; inputting the original sample global channel weights, the positive sample global channel weights and the negative sample global channel weights into a multi-layer perceptron to obtain a processing result; inputting the original sample feature maps, the positive sample feature maps and the negative sample feature maps into the spatial attention layer to obtain spatial local feature weights; performing point multiplication on the original sample feature maps, the positive sample feature maps and the negative sample feature maps and corresponding spatial local feature weights respectively to obtain local enhanced feature maps; wherein the local enhanced feature maps comprise original sample local enhanced feature maps, positive sample local enhanced feature maps and negative sample local enhanced feature maps; performing element subtraction operation on the original sample local enhanced feature maps and the negative sample local enhanced feature maps to obtain local information attention features; calculating a contrast loss function value according to the processing result and the local enhanced feature maps, and calculating an approximation loss function value according to the positive sample local enhanced feature maps and the local information attention features, so as to optimize parameters of the contrast learning module by using the contrast loss function value and the approximation loss function value; performing element addition operation on the global enhanced feature map and the local information attention features to obtain a positive sample global enhanced feature map; wherein the positive sample global enhanced feature map is used for input into the neck network.

[0219] Further, the process that the three-dimensional point cloud processing module generates a corresponding size type detection result for each workpiece in the three-dimensional point cloud comprises: performing point cloud filtering and segmentation operation on the three-dimensional point cloud to obtain a workpiece point cloud of each workpiece; calculating a point cloud plane equation of each workpiece according to the workpiece point cloud, and determining an inclination angle of the point cloud plane equation; determining a minimum bounding box of the workpiece point cloud according to the point cloud plane equation, and calculating a volume of the workpiece point cloud according to the minimum bounding box; determining a size type detection result of each workpiece according to an area of the workpiece point cloud, the volume of the workpiece point cloud and the inclination angle of the point cloud plane equation.

[0220] Further, the painting production line is provided with a workpiece hanging rack, a three-dimensional laser camera and a plurality of two-dimensional cameras, the workpiece hanging area is an area below the workpiece hanging rack, all the two-dimensional cameras are in the same plane, and the shooting direction of the three-dimensional laser camera is perpendicular to the shooting direction of each two-dimensional camera.

[0221] Correspondingly, the process that the two-dimensional image processing module acquires the two-dimensional image of the workpiece hanging area of the coating production line includes: acquiring partial images of the workpiece hanging area of the coating production line from multiple directions by using all the two-dimensional cameras, and splicing the partial images to obtain the two-dimensional image.

[0222] Correspondingly, the process that the three-dimensional point cloud processing module acquires the three-dimensional point cloud of the workpiece hanging area includes: controlling the three-dimensional laser camera to move to a preset position in sequence, and acquiring a partial point cloud of the workpiece hanging area at the preset position; and splicing the partial point cloud to obtain the three-dimensional point cloud of the workpiece hanging area.

[0223] Further, the system further includes a temperature prediction model, which is configured to determine a mapping relationship between workpiece total mass and baking temperature according to historical data before determining a corresponding target baking temperature according to the total mass of the current batch of workpieces, establish a temperature prediction model according to the mapping relationship, and optimize the temperature prediction model by using a gradient descent algorithm.

[0224] Correspondingly, the process that the temperature control module determines a corresponding target baking temperature according to the total mass of the current batch of workpieces, and controls the coating production line to perform a corresponding baking operation according to the target baking temperature includes: inputting the total mass of the current batch of workpieces into the temperature prediction model to obtain a target baking temperature; and adjusting the opening degree of a natural gas adjusting valve according to the target baking temperature, so as to control a hot air fan to perform a baking operation on the workpieces in the coating production line according to the target baking temperature.

[0225] Since the embodiments of the system part correspond to the embodiments of the method part, the embodiments of the system part are described in the description of the embodiments of the method part, and are not described here.

[0226] The application further provides a storage medium having a computer program stored thereon, and the computer program can implement the steps provided in the above embodiments when executed. The storage medium can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0227] The application further provides an electronic device, which can include a memory and a processor, the memory has a computer program stored therein, and the processor can implement the steps provided in the above embodiments when calling the computer program in the memory. Of course, the electronic device can also include various network interfaces, power supplies and other components.

[0228] The various embodiments described in the specification are intended to be exemplary only. The scope of the application is therefore intended to be limited solely by the scope of the appended claims. It will be understood by those within the art that, in light of the wide variety of embodiments disclosed herein, the present application is capable of changes in the details of implementation and application. It will be understood by those within the art that the application is capable of changes in the details of implementation and application without departing from the scope of the application. It will be understood that any reference to the application includes a reference to any such variations. Moreover, it will be understood that the application is capable of out-of-the-box implementation where the application is implemented according to the embodiments disclosed herein without further modification. It will be understood that any reference to the application includes a reference to any such out-of-the-box implementation. It will be understood that any reference to the application includes a reference to any such out-of-the-box implementation.

[0229] It is also important to note that the use of relational terms such as first and second, and the like, are used solely to distinguish one from another entity or action without necessarily implying any actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, preclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

Claims

1. A method of adjusting the baking temperature of a painting line, characterized in that, The method comprises the following steps: acquiring a two-dimensional image of a workpiece hanging area of a coating production line, and using a target detection algorithm model to generate a workpiece category detection result of each workpiece in the two-dimensional image; acquiring a three-dimensional point cloud of the workpiece hanging area, and generating a corresponding size type detection result for each workpiece in the three-dimensional point cloud; determining the total quality of the workpieces in the batch according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area; determining a corresponding target baking temperature according to the total quality of the workpieces in the batch, and controlling the coating production line to perform a corresponding baking operation according to the target baking temperature; wherein the target detection algorithm model is a YOLOv7-CSM model obtained by inserting a contrastive learning module CSM into a YOLOv7 model; the YOLOv7-CSM model comprises a backbone network, a neck network, a detection head, the contrastive learning module, and a loss calculation module; the contrastive learning module comprises a global average pooling layer, a multilayer perceptron, and a spatial attention layer, and is used to convert a feature map input by the backbone network into a positive sample global enhanced feature map; wherein the size type detection result of each workpiece in the three-dimensional point cloud is generated by performing the following steps: performing point cloud filtering and segmentation operations on the three-dimensional point cloud to obtain a workpiece point cloud of each workpiece; calculating a point cloud plane equation of each workpiece according to the workpiece point cloud, and determining an inclination angle of the point cloud plane equation; determining a minimum bounding box of the workpiece point cloud according to the point cloud plane equation, and calculating the volume of the workpiece point cloud according to the minimum bounding box; determining the size type detection result of each workpiece according to the area of the workpiece point cloud, the volume of the workpiece point cloud, and the inclination angle of the point cloud plane equation.

2. The baking temperature adjusting method for a coating line according to claim 1, wherein The total quality of the workpieces in the batch is determined according to the workpiece category detection result and / or the size type detection result of each workpiece in the workpiece hanging area by performing the following steps: selecting a target workpiece from the workpiece hanging area; querying a size type corresponding to the workpiece category detection result of the target workpiece in a database to obtain a size type query result; determining whether the size type detection result of the target workpiece is the same as the size type query result; if yes, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type detection result; if no, determining a two-dimensional detection confidence and a three-dimensional detection confidence of the target workpiece; wherein the two-dimensional detection confidence is the confidence of the workpiece category detection result, and the three-dimensional detection confidence is the confidence of the size type detection result; if the two-dimensional detection confidence is greater than the three-dimensional detection confidence, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type query result; if the two-dimensional detection confidence is less than the three-dimensional detection confidence, determining the workpiece quality of the target workpiece according to the size type detection result; if the two-dimensional detection confidence is equal to the three-dimensional detection confidence, determining the workpiece quality of the target workpiece according to the workpiece category detection result and the size type detection result. determining whether all workpieces in the workpiece hanging area are selected; if yes, calculating the total mass of the batch of workpieces according to the workpiece mass of all workpieces in the workpiece hanging area; if no, proceeding to the step of selecting a target workpiece from the workpiece hanging area.

3. The baking temperature adjusting method for a coating line according to claim 2, wherein After the target workpiece is selected from the workpiece hanging area, further comprising: if the target workpiece has a corresponding workpiece category detection result and does not have a corresponding size type detection result, determining the workpiece mass of the target workpiece according to the workpiece category detection result and the size type query result; if the target workpiece has a corresponding size type detection result and does not have a corresponding workpiece category detection result, determining the workpiece mass of the target workpiece according to the size type detection result.

4. The baking temperature adjusting method for a coating line according to claim 2, wherein determining the workpiece mass of the target workpiece according to the size type detection result, comprising: querying alternative workpieces matching the size type detection result in a database; setting the average mass of all alternative workpieces as the workpiece mass of the target workpiece.

5. The method of claim 1, wherein the baking temperature is adjusted by a temperature sensor installed in the baking furnace. Before generating the workpiece category detection result of each workpiece in the two-dimensional image by using the target detection algorithm model, further comprising: obtaining training samples from a data set; wherein each training sample includes an original sample, a positive sample and a negative sample, the original sample is an image containing a workpiece, the positive sample is an image within the bounding box of the workpiece in the original sample, and the negative sample is an image outside the bounding box of the workpiece in the original sample; training the target detection algorithm model by using the training samples.

6. The baking temperature adjusting method for a coating line according to claim 5, wherein The process of converting the feature map input by the skeleton network into the positive sample global enhanced feature map by the contrast learning module comprises: receiving the original sample feature map, the positive sample feature map and the negative sample feature map output by the skeleton network; inputting the original sample feature map, the positive sample feature map and the negative sample feature map into the global average pooling layer to obtain global channel weights; wherein the global channel weights include original sample global channel weights, positive sample global channel weights and negative sample global channel weights; point-multiplying the original sample feature map and the original sample global channel weights to obtain a global enhanced feature map; inputting the original sample global channel weights, the positive sample global channel weights and the negative sample global channel weights into a multilayer perceptron to obtain a processing result; inputting the original sample feature map, the positive sample feature map and the negative sample feature map into the spatial attention layer to obtain spatial local feature weights; point-multiplying the original sample feature map, the positive sample feature map and the negative sample feature map with the corresponding spatial local feature weights respectively to obtain local enhanced feature maps; wherein the local enhanced feature maps include an original sample local enhanced feature map, a positive sample local enhanced feature map and a negative sample local enhanced feature map; performing element subtraction operation on the original sample local enhanced feature map and the negative sample local enhanced feature map to obtain a local information attention feature; According to the processing result and the local enhanced feature map, a contrast loss function value is calculated, and according to the positive sample local enhanced feature map and the local information attention feature, an approximation loss function value is calculated, so as to optimize the parameters of the contrast learning module by using the contrast loss function value and the approximation loss function value. An element addition operation is performed on the global enhanced feature map and the local information attention feature to obtain a positive sample global enhanced feature map; wherein the positive sample global enhanced feature map is used as input to the neck network.

7. The method of claim 1, wherein the baking temperature is adjusted to a temperature of 80 to 100°C. The coating production line is provided with a workpiece hanging rack, a three-dimensional laser camera and a plurality of two-dimensional cameras, the workpiece hanging area is an area below the workpiece hanging rack, all the two-dimensional cameras are in the same plane, and the shooting direction of the three-dimensional laser camera is perpendicular to the shooting direction of each two-dimensional camera. Correspondingly, acquiring a two-dimensional image of a workpiece hanging area of a coating production line comprises: Capturing local images of the workpiece hanging area of the coating production line from multiple directions using all the two-dimensional cameras, and splicing the local images to obtain the two-dimensional image; Correspondingly, acquiring a three-dimensional point cloud of the workpiece hanging area comprises: Controlling the three-dimensional laser camera to move to a preset position in sequence and acquiring local point clouds of the workpiece hanging area at the preset positions; Splicing the local point clouds to obtain the three-dimensional point cloud of the workpiece hanging area.

8. The method of claim 1, wherein the baking temperature is adjusted to a temperature of 80 to 100°C. Before determining the corresponding target baking temperature according to the total mass of the workpieces in the current batch, further comprising: Determining a mapping relationship between the total mass of the workpieces and the baking temperature according to historical data, establishing a temperature prediction model according to the mapping relationship, and optimizing the temperature prediction model using a gradient descent algorithm; Correspondingly, determining the corresponding target baking temperature according to the total mass of the workpieces in the current batch and controlling the coating production line to perform corresponding baking operations according to the target baking temperature comprises: Inputting the total mass of the workpieces in the current batch into the temperature prediction model to obtain the target baking temperature; Adjusting the opening degree of the natural gas regulating valve according to the target baking temperature, so as to control the hot air fan to perform baking operations on the workpieces in the coating production line according to the target baking temperature.

9. A bake temperature regulation system for a finishing line, characterized by, Comprise: A two-dimensional image processing module is configured to acquire a two-dimensional image of a workpiece hanging area of a coating production line and generate workpiece category detection results for each workpiece in the two-dimensional image using a target detection algorithm model; A three-dimensional point cloud processing module is configured to acquire a three-dimensional point cloud of the workpiece hanging area and generate corresponding size type detection results for each workpiece in the three-dimensional point cloud; A quality calculation module is configured to determine the total mass of the workpieces in the current batch according to the workpiece category detection results and / or size type detection results for each workpiece in the workpiece hanging area; A temperature control module is configured to determine the corresponding target baking temperature according to the total mass of the workpieces in the current batch and control the coating production line to perform corresponding baking operations according to the target baking temperature. The target detection algorithm model is a YOLOv7-CSM model obtained by inserting a contrast learning module CSM into a YOLOv7 model; the YOLOv7-CSM model comprises a skeleton network, a neck network, a detection head, the contrast learning module and a loss calculation module, the contrast learning module comprises a global average pooling layer, a multilayer perceptron and a spatial attention layer, and the contrast learning module is configured to convert a feature map input by the skeleton network into a positive sample global enhanced feature map; The three-dimensional point cloud processing module generates a corresponding size type detection result for each workpiece in the three-dimensional point cloud, including: performing point cloud filtering and segmentation operations on the three-dimensional point cloud to obtain a workpiece point cloud of each workpiece; calculating a point cloud plane equation of each workpiece according to the workpiece point cloud, and determining an inclination angle of the point cloud plane equation; determining a minimum bounding box of the workpiece point cloud according to the point cloud plane equation, and calculating the volume of the workpiece point cloud according to the minimum bounding box; and determining the size type detection result of each workpiece according to the area of the workpiece point cloud, the volume of the workpiece point cloud and the inclination angle of the point cloud plane equation.

10. An electronic device, comprising: The coating production line comprises a memory and a processor, the memory stores a computer program, and the processor calls the computer program in the memory to realize the steps of the baking temperature adjustment method of the coating production line according to any one of claims 1 to 8.

11. A storage medium, characterized by The storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by the processor to realize the steps of the baking temperature adjustment method of the coating production line according to any one of claims 1 to 8.

Images