Adhesive tape pasting position positioning method, control device and equipment

Abstract

The embodiment of the invention provides an adhesive tape pasting position positioning method, a control device and equipment, and the method comprises the steps: S1, collecting a to-be-pasted region picture, and making a data set; s2, using the data set to train and improve a NanoDet model, and obtaining a detection model, the improvement measures including texture filter and feature enhancement; s3, detecting an input to-be-pasted area by using the detection model, and obtaining position information of the to-be-pasted area; and S4, adhesive tape pasting is carried out according to the position information. The method can adapt to different working scenes, the positioning accuracy of the adhesive tape pasting position is improved, and the product quality is improved.

Description

Technical Field

[0001] The present application relates to the field of positioning, and in particular to a method, control device and equipment for positioning a tape sticking position. Background Art

[0002] Tape application is a key step in industrial automation, directly affecting product quality, production efficiency, and cost control. In areas such as electronic component packaging, automobile manufacturing, and precision instrument assembly, tape application accuracy requirements typically reach submillimeter levels. Traditional methods rely on manual operation or simple visual inspection, which presents problems such as low efficiency, large errors, and high rework rates. Existing technologies mainly include traditional visual positioning based on edge detection, detection models based on deep learning, and multi-sensor fusion solutions. However, these methods face defects such as poor environmental adaptability, insufficient response to geometric deformation, and a conflict between real-time performance and accuracy in complex scenarios. For example, positioning errors increase significantly in strong light or low-contrast environments, and fitting accuracy on curved surfaces or flexible materials is difficult to guarantee.

[0003] Therefore, proposing a tape pasting position positioning method has important practical significance for improving tape pasting efficiency and improving product quality. Summary of the Invention

[0004] This application provides a method, control device and equipment for locating the position of tape sticking, which can improve the positioning accuracy of the tape sticking position by improving the texture filtering and feature enhancement of the NanoDet model.

[0005] In a first aspect, a method for locating a tape sticking position is provided, the method comprising:

[0006] S1: Collect images of the area to be pasted and create a data set;

[0007] S2: Use the dataset to train and improve the NanoDet model to obtain the detection model. The improvement measures include: texture filter and feature enhancement;

[0008] S3: Use the detection model to detect the input area to be pasted and obtain the position information of the area to be pasted;

[0009] S4: Applying tape according to the position information.

[0010] It should be understood that this method is based on deep learning and machine vision technology. By performing texture filtering and feature enhancement on the NanoDet model, it can adapt to various scenarios such as industrial production and logistics packaging, and can improve the accuracy of tape pasting and enhance product quality.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, step S1 includes:

[0012] S101: Collect photos of the area to be pasted at different angles and under different lighting conditions;

[0013] S102: Preprocess the images collected in S101. The preprocessing includes: using image enhancement technology to improve the image quality and scaling the image to a size suitable for NanoDet model input;

[0014] S103: Annotate the preprocessed images and save the annotation information in a format that meets the training requirements of the NanoDet model to complete the production of the dataset.

[0015] In combination with the first aspect, in certain implementations of the first aspect, a texture filter is used to remove textures irrelevant to the area to be pasted, and the texture filter is designed based on a Gaussian difference filter.

[0016] It should be understood that the Gaussian difference filter is constructed based on the principle of Gaussian function. It uses two Gaussian kernels of different scales to perform convolution operations on the input image respectively, and then subtracts the two convolution results, thereby highlighting the edges and detail information in the image and effectively suppressing high-frequency texture noise that is not related to the features of the area to be pasted.

[0017] In combination with the first aspect, in some implementations of the first aspect, the texture filter is jointly optimized with the backbone network of the NanoDet model through learnable weights.

[0018] It should be understood that jointly optimizing the learnable weights of the texture filter with the backbone network of the NanoDet model enables better synergy between the filter and the backbone network. Through joint optimization, the parameter adjustment of the texture filter will interact with the parameter update of the backbone network, allowing the filter to provide the subsequent backbone network with input that is more conducive to target detection. The feedback from the backbone network can also prompt the filter to continuously optimize its own parameters, thereby improving the performance of the entire model.

[0019] In combination with the first aspect, in some implementations of the first aspect, the texture filter is located before the feature extraction module of the NanoDet model.

[0020] It's important to understand that when an image enters the feature extraction module, it often contains a variety of complex texture information, some of which may be irrelevant to the target being detected and represent interference. Using a texture filter beforehand can effectively remove this interference, resulting in a cleaner image and more prominent features before entering the feature extraction module. This allows the feature extraction module to more closely focus on extracting features relevant to the target, improving both the efficiency and quality of feature extraction, and ultimately, the performance of the entire model.

[0021] In combination with the first aspect, in some implementations of the first aspect, feature enhancement includes: introducing deformable convolution into the NanoDet model.

[0022] In combination with the first aspect, in some implementations of the first aspect, the deformable convolution is located at the second feature extraction position in the feature extraction module of the NanoDet model, and the offset of the deformable convolution is dynamically generated by the feature map.

[0023] It should be understood that deformable convolution is a convolution method developed based on traditional convolution, which aims to better adapt to the geometric changes of the target. Traditional convolution samples at fixed regular grid positions and has limited adaptability to changes in the shape and posture of the target. Deformable convolution introduces the concept of offset, and learns the offset of each sampling point through an additional convolution layer, so that the sampling points of the convolution kernel can be flexibly offset in space and are no longer limited to fixed positions. The area to be pasted often deforms, so deformable convolution is suitable for the work scenarios faced by this application.

[0024] It should be understood that this application uses feature maps to generate offsets, and deformable convolution can adaptively adjust the positions of sampling points according to information such as the shape and position of the target in the input feature map to better capture the characteristics of the target.

[0025] It should also be understood that the feature extraction module of the NanoDet model includes three feature extraction operations, and this application chooses to introduce deformable convolution during its second feature extraction. The shallow convolution layer (the first feature extraction) mainly extracts some basic and low-level features. At this time, the specific shape and posture information of the target are not obvious, and the advantages of deformable convolution are difficult to give full play to; while the deep convolution layer (the third feature extraction) extracts highly abstract semantic features. At this time, adjusting the sampling position may destroy the abstract representation that has been learned. Therefore, the introduction of deformable convolution during the second feature extraction of this application can better capture the shape and posture changes of the target at this layer, adaptively adjust the sampling position to extract more representative features, and improve the accuracy of the tape pasting position.

[0026] In combination with the first aspect, in certain implementations of the first aspect, step S4 includes: using a visual servo controller to perform tape pasting according to position information and make real-time adjustments.

[0027] In a second aspect, a control device is provided, which includes a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that the method in any one of the implementations in the first aspect is executed.

[0028] In a third aspect, a device for positioning a tape pasting position is provided, the device comprising the control device according to the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A flowchart of a method for locating a tape sticking position provided in an embodiment of the present application.

[0030] Figure 2 This is a flow chart of a method for producing a dataset of locations where tape is to be pasted, provided in an embodiment of the present application.

[0031] Figure 3 A schematic diagram of the improved NanoDet model structure provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The terms used in the following embodiments are only for the purpose of describing specific embodiments and are not intended to be limiting of the present application. As used in the specification of the present application and the appended claims, the singular expressions "a", "an", "", "above", "the" and "this" are intended to also include expressions such as "one or more", unless the context clearly indicates otherwise. It should also be understood that in the following embodiments of the present application, "at least one", "one or more" refer to one, two or more. The term "and / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist; for example, A and / or B may represent: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B may be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship.

[0033] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in yet other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0034] Taping is a critical step in industrial automation, directly impacting product quality, production efficiency, and cost control. In fields such as electronic component packaging, automotive manufacturing, and precision instrument assembly, tape application often requires submillimeter precision. Traditional methods rely on manual operation or simple visual inspection, resulting in low efficiency, large errors, and high rework rates.

[0035] The present invention provides a method, control device, and apparatus for locating the location of tape application. Based on deep learning and machine vision technology, this method incorporates a texture filter into the NanoDet model and simultaneously enhances the NanoDet model's features, effectively improving the accuracy of tape application location positioning and being applicable to a variety of work scenarios.

[0036] Figure 1 This is a flowchart of a method for locating a tape attachment position provided in an embodiment of the present application. In some examples, the method includes:

[0037] S1: Collect images of the area to be pasted and create a data set;

[0038] S2: Use the dataset to train and improve the NanoDet model to obtain the detection model. The improvement measures include: texture filter and feature enhancement;

[0039] S3: Use the detection model to detect the input area to be pasted and obtain the position information of the area to be pasted;

[0040] S4: Applying tape according to the position information.

[0041] Figure 2 This is a flow chart of a method for generating a dataset of locations where tape is to be pasted, provided in an embodiment of the present application. In some examples, the method includes:

[0042] S101: Collect photos of the area to be pasted at different angles and under different lighting conditions;

[0043] S102: Preprocess the images collected in S101. The preprocessing includes: using image enhancement technology to improve the image quality and scaling the image to a size suitable for NanoDet model input;

[0044] S103: Annotate the preprocessed images and save the annotation information in a format that meets the training requirements of the NanoDet model to complete the production of the dataset.

[0045] In one possible implementation, in addition to using common histogram equalization to adjust the contrast of the image, the image enhancement technology also uses adaptive histogram equalization to better handle contrast changes in local areas of the image, making the details of the area to be pasted clearer.

[0046] In one possible implementation, a combination of median filtering and Gaussian filtering is used to reduce noise interference in the image. First, median filtering is used to remove salt and pepper noise in the image, and then Gaussian filtering is used to smooth the image to remove Gaussian noise and further improve the image quality.

[0047] In one possible implementation, a bicubic interpolation algorithm is used to scale the image to a size suitable for the NanoDet model input to ensure the smoothness of the edges and the integrity of the details of the image during the scaling process, avoiding jagged or blurring phenomena.

[0048] In some examples, a texture filter is used to remove textures irrelevant to the area to be pasted, and the texture filter is designed based on a Gaussian difference filter.

[0049] In some examples, the texture filter is jointly optimized with the backbone network of the NanoDet model via learnable weights.

[0050] In some examples, the texture filter precedes the feature extraction module of the NanoDet model.

[0051] In one possible implementation, the texture filter is implemented by generating two Gaussian kernels of different scales based on a Gaussian function in a custom convolution layer; performing a convolution operation on the input image using the two Gaussian kernels to obtain two smoothed images of different scales; and subtracting the two smoothed images to obtain a Gaussian difference filtering result, which is the result of the texture filtering.

[0052] In one possible implementation, to better adapt the Difference of Gaussian filter to different datasets and tasks, the parameters of the Gaussian kernel can be set as learnable parameters and jointly optimized with the NanoDet model backbone network. During training, these parameters are automatically adjusted through the backpropagation algorithm, allowing the filter to learn the optimal filtering effect.

[0053] In some examples, feature enhancement includes introducing deformable convolutions in the NanoDet model.

[0054] In some examples, the deformable convolution is located at the second feature extraction position in the feature extraction module of the NanoDet model, and the offset of the deformable convolution is dynamically generated by the feature map.

[0055] In some examples, step S4 includes: using a visual servo controller to perform tape sticking according to position information and make real-time adjustments.

[0056] An embodiment of the present application provides a control device, which includes a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that any one of the methods described in the aforementioned specific embodiments is executed.

[0057] An embodiment of the present application also provides a tape sticking position positioning device, which includes the control device as described above.

[0058] The above are only preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by ordinary technicians in this field based on the contents disclosed in the present invention should be included in the protection scope recorded in the claims.

Claims

1. A method for locating a tape sticking position, characterized in that: The method comprises: S1: Collect images of the area to be pasted and create a data set; S2: Using the dataset to train and improve the NanoDet model to obtain a detection model, where the improvement measures include: texture filter and feature enhancement; S3: Detecting the input area to be pasted using the detection model to obtain position information of the area to be pasted; S4: performing tape pasting according to the position information.

2. The method according to claim 1, characterized in that The step S1 comprises: S101: Collect photos of the area to be pasted at different angles and under different lighting conditions; S102: Preprocessing the image collected in S101, wherein the preprocessing includes: using image enhancement technology to improve the quality of the image and scaling the image to a size suitable for the NanoDet model input; S103: Annotate the preprocessed image, save the annotation information in a format that meets the training requirements of the NanoDet model, and complete the production of the data set.

3. The method according to claim 1, characterized in that The texture filter is used to remove textures irrelevant to the area to be pasted, and the texture filter is designed based on a Gaussian difference filter.

4. The method according to claim 3, characterized in that The texture filter is jointly optimized with the backbone network of the NanoDet model through learnable weights.

5. The method according to claim 4, characterized in that The texture filter is located before the feature extraction module of the NanoDet model.

6. The method according to claim 1, wherein The feature enhancement includes: introducing deformable convolution into the NanoDet model.

7. The method according to claim 6, characterized in that The deformable convolution is located at the second feature extraction position in the feature extraction module of the NanoDet model, and the offset of the deformable convolution is dynamically generated by the feature map.

8. The method according to claim 1, characterized in that The step S4 includes: using a visual servo controller to perform tape pasting according to the position information and make real-time adjustments.

9. A control device, characterized in that: The method comprises a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that the method according to any one of claims 1 to 8 is performed.

10. A tape sticking position positioning device, characterized in that: The apparatus comprises a control device as claimed in claim 9.

Images