Welding groove identification method and system based on YFD network model

Abstract

The invention relates to the technical field of welding production, in particular to a welding groove recognition method and system based on a YFD network model. The method disclosed by the invention comprises the following steps: step 1, acquiring a welding groove image to be identified; and 2, processing a welding groove image to be identified by adopting the trained YFD network model so as to identify a groove type. According to the invention, a YFD network model is designed, the feature extraction part is used for carrying out multi-scale feature extraction on the identified welding groove image, the feature fusion part is used for carrying out multi-scale feature fusion, and the detection output part is used for carrying out conversion to obtain three detection heads so as to identify the groove type in the welding groove image to be identified. The method has the characteristics of high recognition precision and high processing speed, and is particularly suitable for actual welding production.

Description

Technical Field

[0001] The present invention relates to the technical field of welding production, and more specifically to: 1. a welding groove recognition method based on a YFD network model; 2. a welding groove recognition system based on a YFD network model. Background Art

[0002] With the development of deep learning and machine vision technology, they are increasingly used in the fields of weld tracking, identification, and detection, but the identification of the weld groove is less involved. This is because welding is mostly manual and there is no need to identify the weld groove.

[0003] With the introduction of welding robots in welding production, the adjustment of welding parameters will waste a lot of time. The existing inertial parameter adjustment method is still manually adjusted based on experience. However, the inventor believes that it is necessary to correctly identify the welding groove of the welding workpiece in order to use appropriate welding parameters to ensure the welding effect. Therefore, it is of practical significance to design a welding groove recognition method that can ensure its recognition accuracy and recognition speed. Summary of the invention

[0004] Based on this, it is necessary to address the problem of the lack of welding groove recognition methods in the existing market, and provide a welding groove recognition method and system based on the YFD network model.

[0005] The present invention is implemented by the following technical solutions:

[0006] In a first aspect, the present invention discloses a welding groove recognition method based on a YFD network model, comprising the following steps:

[0007] Step 1, obtaining a welding groove image to be identified;

[0008] Step 2: Use the trained YFD network model to process the welding groove image to be identified to identify the groove type.

[0009] Among them, the YFD network model includes: feature extraction part, feature fusion part, and detection output part.

[0010] The feature extraction unit is used to perform multi-scale feature extraction on the identified welding groove image to obtain feature extraction images P4, P6, and P9 of three scales; the specifications of P4, P6, and P9 decrease in sequence.

[0011] The feature fusion unit is used to perform multi-scale feature fusion on P4, P6, and P9 to obtain feature fusion maps T1, T2, and T3 of three scales; the specifications of T1, T2, and T3 decrease in sequence.

[0012] The detection output part is used to convert T1, T2, and T3 to obtain three detection heads, so as to identify the groove type in the welding groove image to be identified.

[0013] This welding groove recognition method based on the YFD network model implements the method or process according to the embodiment of the present disclosure.

[0014] In a second aspect, the present invention discloses a welding groove recognition system based on a YFD network model, which uses the welding groove recognition method based on a YFD network model disclosed in the first aspect.

[0015] The welding groove recognition system based on the YFD network model includes: an image acquisition module and a welding groove recognition module.

[0016] The image acquisition module is used to acquire the welding groove image to be identified.

[0017] The welding groove recognition module is used to process the welding groove image to be identified using the trained YFD network model to identify the groove type.

[0018] The welding groove recognition system based on the YFD network model implements the method or process according to the embodiment of the present disclosure.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. The present invention designs a YFD network model, which is used to perform multi-scale feature extraction on the identified welding groove image through a feature extraction unit, is used to perform multi-scale feature fusion through a feature fusion unit, and is converted through a detection output unit to obtain three detection heads to identify the groove type in the welding groove image to be identified; the present invention has the characteristics of high recognition accuracy and fast processing speed, and is particularly suitable for actual welding production use.

[0021] 2. The YFD network model of the present invention introduces the FasterNet layer in the feature extraction part, which can reduce redundant calculations and improve the running speed.

[0022] 3. The YFD network model of the present invention introduces a DAttention layer in the feature fusion part, so that the feature fusion will focus on the groove area, and at the same time help the model to more effectively fuse features between different levels, which can reduce the amount of calculation while maintaining good performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1 A data flow diagram of a welding groove identification method based on a YFD network model provided in Example 1 of the present invention;

[0025] Figure 2 for Figure 1 The structure diagram of the FasterNet layer in the figure;

[0026] Figure 3 for Figure 1 The structure diagram of the C2f layer;

[0027] Figure 4 This is a simulation result diagram provided for Example 1 of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component at the same time. When a component is considered to be "fixed to" another component, it may be directly fixed on the other component or there may be a central component at the same time.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0031] Example 1

[0032] See also Figure 1 , Figure 1A data flow diagram of a welding groove recognition method based on a YFD network model is shown, which actually also shows a flow chart of a welding groove recognition method based on a YFD network model, which includes:

[0033] Step 1: Obtain the welding groove image to be identified.

[0034] For example, a photographing device (such as a CCD industrial camera, etc.) may be used to photograph the welding groove to obtain the welding groove image to be identified; or the welding groove image to be identified may be directly acquired from a set of welding groove images to be identified.

[0035] Step 2: Use the trained YFD network model to process the welding groove image to be identified to identify the groove type.

[0036] The core of the present invention is to provide a new structure of welding groove recognition model - YFD network model. It should be noted that the present invention needs to use a trained YFD network model, and its network parameters are in an optimal state. As for the training process of the YFD network model, reference can be made to the conventional training method of the neural network, which will not be repeated here.

[0037] See also Figure 1 ,From the functional point of view, the YFD network model includes : feature ,extraction part, feature fusion part, and detection output part.

[0038] The following describes them one by one:

[0039] 1. The feature extraction unit is used to perform multi-scale feature extraction on the identified welding groove image to obtain feature extraction images P4, P6, and P9 of three scales; the specifications of P4, P6, and P9 decrease in sequence.

[0040] like Figure 1 As shown, the feature extraction unit includes: 4 Merging layers, 4 FasterNet layers, and 1 SPPF layer.

[0041] In the feature extraction section:

[0042] The first Merging layer is used to downsample the input of the feature extraction unit;

[0043] The first FasterNet layer is used to partially filter and extract features from the output of the first Merging layer;

[0044] The second Merging layer is used to downsample the output of the first FasterNet layer;

[0045] The second FasterNet layer is used to partially filter and extract features from the output of the second Merging layer to obtain P4;

[0046] The third Merging layer is used to downsample the output of the second FasterNet layer;

[0047] The third FasterNet layer is used to partially filter and extract features from the output of the third Merging layer to obtain P6;

[0048] The fourth Merging layer is used to downsample the output of the third FasterNet layer;

[0049] The fourth FasterNet layer is used to partially filter and extract features from the output of the fourth Merging layer;

[0050] The SPPF layer is used to perform pooling on the output of the 4th FasterNet layer to obtain P9.

[0051] For the feature extraction part, it should be noted that:

[0052] ① The four Merging layers have the same structure, including: 1 Conv2d layer and 1 BN layer.

[0053] The Merging layer downsamples and expands the number of channels in the network: Downsampling is used to reduce the resolution of the input image in order to unify the number of input channels of subsequent network layers, which can enhance the robustness and feature extraction capabilities of the model.

[0054] For details, see Figure 1 , in a Merging layer:

[0055] The Conv2d layer is used to perform a two-dimensional convolution operation on the input of the Merging layer;

[0056] The BN layer is used to perform batch normalization on the output of the Conv2d layer to obtain the output of the Merging layer.

[0057] ②The four FasterNet layers have the same structure, including: 1 PConv layer, 2 Conv2d layers, 1 BN layer, 1 Relu activation function layer, and 1 overlay layer.

[0058] The FasterNet layer reduces the model's inference time by optimizing the calculation path and lightweighting the structure of the feature extraction part, aiming to enable the neural network to process data and make predictions faster without compromising the accuracy of the model.

[0059] For details, see Figure 2 , in a FasterNet layer:

[0060] The PConv layer is used to perform partial convolution on the input of the FasterNet layer (the most significant difference between the PConv layer and the Conv layer is that conventional convolution is applied to only a part of the input channels to extract features, while the other channels remain unchanged; this not only reduces the amount of calculation but also effectively extracts features);

[0061] The first Conv2d layer is used to perform a two-dimensional convolution operation on the output of the PConv layer;

[0062] The BN layer is used to perform batch normalization on the output of the first Conv2d layer;

[0063] The Relu activation function layer is used to activate the output of the BN layer using the Relu activation function;

[0064] The second Conv2d layer is used to perform a two-dimensional convolution operation on the output of the BN layer;

[0065] The superposition layer is used to superimpose the input of the FasterNet layer and the output of the second Conv2d layer to obtain the output of the FasterNet layer.

[0066] It should be noted that, in the first embodiment, the specification of P4 is set to 80 pixels×80 pixels, the specification of P6 is set to 40 pixels×40 pixels, and the specification of P9 is set to 20 pixels×20 pixels.

[0067] 2. The feature fusion unit is used to perform multi-scale feature fusion on P4, P6, and P9 to obtain feature fusion maps T1, T2, and T3 of three scales; the specifications of T1, T2, and T3 decrease in sequence.

[0068] like Figure 1 As shown, the feature fusion unit includes: 2 upstream aggregation layers, 4 C2f layers, 2 Conv layers, 2 Concat layers, and 2 DAttention layers.

[0069] In the feature fusion section:

[0070] The first uplink aggregation layer is used to perform uplink aggregation based on P9 and P6 to obtain the fused feature map F1;

[0071] The first C2f layer is used to aggregate the features of multi-scale information of F1;

[0072] The second uplink aggregation layer is used to perform uplink aggregation based on the output of the first C2f layer and P4;

[0073] The second C2f layer is used to perform feature aggregation of multi-scale information on the output of the second uplink aggregation layer to obtain T1;

[0074] The first Conv layer is used to perform convolution on T1;

[0075] The first Concat layer is used to perform horizontal fusion of the output of the first Conv layer and the output of the first C2f layer to obtain the fused feature map F2;

[0076] The first DAttention layer is used to focus attention on F2;

[0077] The third C2f layer is used to aggregate the features of multi-scale information on the output of the first DAttention layer to obtain T2;

[0078] The second Conv layer is used to perform convolution on T2;

[0079] The second Concat layer is used to perform horizontal fusion of the output of the second Conv layer and P9 to obtain the fused feature map F3;

[0080] The second DAttention layer is used to focus attention on F3;

[0081] The fourth C2f layer is used to aggregate the features of multi-scale information on the output of the second DAttention layer to obtain T3.

[0082] For the feature fusion part, it should be noted that:

[0083] ① The two uplink aggregation layers have the same structure, both including: 1 Upsample layer and 1 Concat layer.

[0084] Specifically,

[0085] (1) In the first uplink aggregation layer:

[0086] The Upsample layer is used to upsample P9 to the same specifications as P6;

[0087] The Concat layer is used to fuse the output of the Upsample layer with P6 at the same level;

[0088] (2) In the second uplink aggregation layer:

[0089] The Upsample layer is used to upsample the output of the first C2f layer to the same specifications as P4;

[0090] The Concat layer is used to fuse the output of the Upsample layer with P4 at the same level.

[0091] That is to say, in the feature fusion unit, a feature pyramid network is constructed to achieve multi-scale feature fusion from bottom to top layers, so that the features of each layer contain rich contextual information.

[0092] ②The structures of the 4 C2f layers are the same, including: 2 Conv layers, 1 Split layer, 1 Darknet Bottleneck part, and 1 Concat layer.

[0093] See also Figure 3 , in a C2f layer:

[0094] The first Conv layer is used to perform convolution processing on the input of the C2f layer;

[0095] The Split layer is used to split the output of the first Conv layer into two parts: feature map A1 and feature map A2;

[0096] The DarknetBottleneck part includes at least three DarknetBottleneck layers (three layers are designed in this embodiment 1); wherein the first DarknetBottleneck layer is used to capture complex detail features from the feature map A1; the output of the previous DarknetBottleneck layer is used as the input of the next DarknetBottleneck layer and the input of the Concat layer;

[0097] The Concat layer is used to merge the output of each DarknetBottleneck layer and feature map A2 of the DarknetBottleneck part, so that the model can use more hierarchical information;

[0098] The second Conv layer is used to perform convolution on the output of the Concat layer.

[0099] In other words, the C2f layer obtains richer gradient flow information while ensuring lightweight by running more gradient flow branches in parallel, which can improve the accuracy of subsequent detection.

[0100] ③The structures of the two DAttention layers are the same, which aims to focus the features on the groove area and help the model to more effectively integrate the features between different levels, which facilitates the classification and detection of subsequent detection heads.

[0101] Since the contour of the welding groove has significant image features, the deformable attention mechanism is used to focus only on the groove of the image without fixedly processing the entire image. This can reduce the amount of computation while maintaining good performance.

[0102] It should be emphasized that the DAttention layer is a plug-and-play attention mechanism, but its location is particular - it does not mean that adding it to the model will have a good effect; moreover, the introduction of the DAttention layer will lead to a large increase in the number of parameters of the model; therefore, in this embodiment 1, only two DAttention layers are selected and the above-mentioned placement method is used to achieve the best use effect when the model is not complicated.

[0103] It should be noted that, in the first embodiment, the specification of T1 is set to 80 pixels×80 pixels, the specification of T2 is set to 40 pixels×40 pixels, and the specification of T3 is set to 20 pixels×20 pixels.

[0104] 3. The detection output unit is used to convert T1, T2, and T3 to obtain three detection heads to identify the groove type in the welding groove image to be identified.

[0105] like Figure 1 As shown, the detection output part includes: 3 Detect layers;

[0106] The first Detect layer is used to convert T1 into a detection head for detecting small objects;

[0107] The second Detect layer is used to convert T2 into a detection head for detecting medium-sized objects;

[0108] The third Detect layer is used to transform T3 into a detection head for detecting large objects.

[0109] It should be noted that, in this embodiment 1, the detection head generated by the first Detect layer is used to detect objects with a size of 80 pixels × 80 pixels, which can better capture the details of small targets; the detection head generated by the second Detect layer is used to detect objects with a size of 40 pixels × 40 pixels; the detection head generated by the third Detect layer is used to detect objects with a size of 20 pixels × 20 pixels, which does not require too much detailed information but requires global context information.

[0110] Among them, the image detail increases successively when comparing 20 pixels × 20 pixels, 40 pixels × 40 pixels, and 80 pixels × 80 pixels, so the size of detectable objects becomes smaller successively.

[0111] Then, using the trained YFD network model mentioned above, the groove type can be identified for the welding groove image to be identified.

[0112] Simulation Verification

[0113] This embodiment 1 performs simulation verification on the above-mentioned method:

[0114] The YFD network model (abbreviated as YFD-Net) was tested on the same dataset with the other three existing network models (YOLOv8n, YOLOv5s, and YOLOv8n-DAT).

[0115] Four performance indicators (including: Precision, average precision mAP@0.5, recall rate Recall, F1 score F1_Score) are introduced for evaluation. See Table 1 below for details. Figure 4 .

[0116] Table 1 Comparison results of model experiments

[0117] Model Accuracy Recall F1_Score YOLOv8n 94.13 91.12 92.73 YOLOv5s 93.22 90.11 91.63 YOLOv8n-DAT 94.31 92.27 93.18 YFD-Net 96.82 94.41 95.58

[0118] As can be seen from Table 1, YFD-Net outperforms the other three groups of models in all three performance indicators;

[0119] Depend on Figure 4 It can be seen that compared with the other three groups of models, YFD-Net has a higher mAP@0.5 at the same number of iterations; when the number of iterations reaches 30, the curve fluctuates more gently, showing better model performance.

[0120] Combined with the simulation results, it can be seen that the welding groove recognition method based on the YFD network model proposed in Example 1 has certain advantages.

[0121] Example 2

[0122] This embodiment 2 provides a welding groove recognition system based on the YFD network model, which uses the welding groove recognition method based on the YFD network model disclosed in embodiment 1.

[0123] The welding groove recognition system based on the YFD network model includes: an image acquisition module and a welding groove recognition module.

[0124] The image acquisition module is used to acquire the welding groove image to be identified.

[0125] The welding groove recognition module is used to process the welding groove image to be identified using the trained YFD network model to identify the groove type.

[0126] Since this system uses the welding groove recognition method based on the YFD network model in Example 1, it also has the same effect and will not be repeated here.

[0127] Example 3

[0128] This embodiment 3 discloses a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the welding groove identification method based on the YFD network model disclosed in embodiment 1 are implemented.

[0129] This embodiment 3 also discloses a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the steps of the welding groove identification method based on the YFD network model disclosed in embodiment 1 are executed.

[0130] This embodiment 3 also discloses a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the welding groove identification method based on the YFD network model disclosed in embodiment 1 are implemented.

[0131] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A welding groove recognition method based on YFD network model, characterized in that: The following steps are involved: Step 1, obtaining a welding groove image to be identified; Step 2: Use the trained YFD network model to process the welding groove image to be identified to identify the groove type; Wherein, the YFD network model includes: A feature extraction unit is used to perform multi-scale feature extraction on the identified welding groove image to obtain feature extraction images P4, P6, and P9 of three scales; the specifications of P4, P6, and P9 decrease in sequence; The feature fusion unit is used to perform multi-scale feature fusion on P4, P6, and P9 to obtain feature fusion graphs T1, T2, and T3 of three scales; the specifications of T1, T2, and T3 decrease in sequence; as well as The detection output part is used to convert T1, T2, and T3 to obtain three detection heads so as to identify the groove type in the welding groove image to be identified.

2. The welding groove recognition method based on the YFD network model according to claim 1 is characterized in that: In step one, the welding groove is photographed to obtain a welding groove image to be identified.

3. The welding groove identification method based on the YFD network model according to claim 1 is characterized in that: The feature extraction unit includes: 4 Merging layers, 4 FasterNet layers, and 1 SPPF layer; In the feature extraction section: The first Merging layer is used to downsample the input of the feature extraction unit; The first FasterNet layer is used to partially filter and extract features from the output of the first Merging layer; The second Merging layer is used to downsample the output of the first FasterNet layer; The second FasterNet layer is used to partially filter and extract features from the output of the second Merging layer to obtain P4; The third Merging layer is used to downsample the output of the second FasterNet layer; The third FasterNet layer is used to partially filter and extract features from the output of the third Merging layer to obtain P6; The fourth Merging layer is used to downsample the output of the third FasterNet layer; The fourth FasterNet layer is used to partially filter and extract features from the output of the fourth Merging layer; The SPPF layer is used to perform pooling on the output of the 4th FasterNet layer to obtain P9; Among them, the specifications of P4 are 80 pixels × 80 pixels, the specifications of P6 are 40 pixels × 40 pixels, and the specifications of P9 are 20 pixels × 20 pixels.

4. The welding groove identification method based on the YFD network model according to claim 3 is characterized in that: The Merging layer includes: 1 Conv2d layer and 1 BN layer; In a merging layer: The Conv2d layer is used to perform a two-dimensional convolution operation on the input of the Merging layer; The BN layer is used to perform batch normalization on the output of the Conv2d layer to obtain the output of the Merging layer.

5. The welding groove recognition method based on the YFD network model according to claim 3 is characterized in that: The FasterNet layer includes: 1 PConv layer, 2 Conv2d layers, 1 BN layer, 1 Relu activation function layer, and 1 overlay layer; In a FasterNet layer: The PConv layer is used to perform partial convolution on the input of the FasterNet layer; The first Conv2d layer is used to perform a two-dimensional convolution operation on the output of the PConv layer; The BN layer is used to perform batch normalization on the output of the first Conv2d layer; The Relu activation function layer is used to activate the output of the BN layer using the Relu activation function; The second Conv2d layer is used to perform a two-dimensional convolution operation on the output of the BN layer; The superposition layer is used to superimpose the input of the FasterNet layer and the output of the second Conv2d layer to obtain the output of the FasterNet layer.

6. The welding groove identification method based on the YFD network model according to claim 1 is characterized in that: The feature fusion unit includes: 2 uplink aggregation layers, 4 C2f layers, 2 Conv layers, 2 Concat layers, and 2 DAttention layers; In the feature fusion section: The first uplink aggregation layer is used to perform uplink aggregation based on P9 and P6 to obtain the fused feature map F1; The first C2f layer is used to aggregate the multi-scale information features of F1; The second uplink aggregation layer is used to perform uplink aggregation based on the output of the first C2f layer and P4; The second C2f layer is used to perform feature aggregation of multi-scale information on the output of the second uplink aggregation layer to obtain T1; The first Conv layer is used to perform convolution on T1; The first Concat layer is used to perform horizontal fusion of the output of the first Conv layer and the output of the first C2f layer to obtain the fused feature map F2; The first DAttention layer is used to focus attention on F2; The third C2f layer is used to aggregate the features of multi-scale information on the output of the first DAttention layer to obtain T2; The second Conv layer is used to perform convolution on T2; The second Concat layer is used to perform horizontal fusion of the output of the second Conv layer and P9 to obtain the fused feature map F3; The second DAttention layer is used to focus attention on F3; The fourth C2f layer is used to aggregate the features of multi-scale information on the output of the second DAttention layer to obtain T3; Among them, the specification of T1 is 80 pixels × 80 pixels, the specification of T2 is 40 pixels × 40 pixels, and the specification of T3 is 20 pixels × 20 pixels.

7. The welding groove recognition method based on the YFD network model according to claim 6 is characterized in that: The uplink aggregation layer includes: 1 Upsample layer and 1 Concat layer; In the first upstream aggregation layer: The Upsample layer is used to upsample P9 to the same specifications as P6; The Concat layer is used to fuse the output of the Upsample layer with P6 at the same level; In the second upstream aggregation layer: The Upsample layer is used to upsample the output of the first C2f layer to the same specifications as P4; The Concat layer is used to fuse the output of the Upsample layer with P4 at the same level.

8. The welding groove recognition method based on the YFD network model according to claim 1 is characterized in that: The detection output unit includes: 3 Detect layers; The first Detect layer is used to convert T1 into a detection head for detecting small objects; The second Detect layer is used to convert T2 into a detection head for detecting medium-sized objects; The third Detect layer is used to transform T3 into a detection head for detecting large objects.

9. A welding groove recognition system based on YFD network model, characterized in that: It uses the welding groove recognition method based on the YFD network model as described in any one of claims 1 to 8; The welding groove recognition system based on the YFD network model includes: An image acquisition module, which is used to acquire the welding groove image to be identified; as well as The welding groove recognition module is used to process the welding groove image to be identified using the trained YFD network model to identify the groove type.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the welding groove identification method based on the YFD network model as described in any one of claims 1 to 8 are implemented.

Images