Ultrasonic edge sealing technology and system for pull-up pants

Abstract

The present application relates to the field of intelligent suturing technology, and specifically discloses an ultrasonic edge-sealing process and system for pull-up pants, which uses artificial intelligence control technology based on deep learning to obtain the side view and top view of the pull-up pants fabric to be welded, and then finally obtains the decoding value through the target detection network, the convolutional neural network of the feature extractor and the decoder, so as to automatically and intelligently recommend the appropriate ultrasonic amplitude value. In this way, the influence of too high or too low amplitude on the welding quality can be avoided in the sewing of pull-up pants, thereby improving the welding efficiency.

Description

Technical Field

[0001] The present application relates to the field of intelligent sewing technology, and more specifically, to an ultrasonic edge sealing process and system for pull-up pants. Background Art

[0002] Pull-ups are diapers for infants or toddlers, usually made of absorbent material. Compared to traditional diapers, pull-ups have a design similar to underwear, so they do not require any fixing devices and are easier to put on and take off.

[0003] The ultrasonic edge-sealing method for pull-up pants is a widely used technology that can firmly bond two fabric edges together without using thread stitching. Generally, the ultrasonic edge-sealing method for pull-up pants can be carried out in the following steps: prepare the pull-up pants material that needs to be edge-sealed and place it on the ultrasonic welding machine; adjust the parameters of the ultrasonic welding machine, including amplitude, time and pressure, to ensure the best effect; align the pull-up pants material with the welding head of the ultrasonic welding machine to ensure a complete fit; start the ultrasonic welding machine, transfer ultrasonic energy to the material, form high temperature at the welding head and melt the material; when the welding head reaches the required time and pressure, stop the ultrasonic welding machine and wait for the material to cool; repeat the above steps until all areas that need edge-sealing are sealed.

[0004] In the above-mentioned step of ultrasonic edge sealing for pull-up pants, the amplitude refers to the maximum displacement of the ultrasonic wave in the material, which determines the high temperature generated by the welding head part and the degree of melting of the material. In practical applications, too high or too low amplitude will affect the welding quality, so an optimized ultrasonic edge sealing method for pull-up pants is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an ultrasonic edge sealing process and system for pull-up pants, which obtains the side view and top view of the pull-up pants fabric to be welded by adopting artificial intelligence control technology based on deep learning, and then finally obtains the decoding value through the target detection network, the convolutional neural network of the feature extractor and the decoder, so as to automatically and intelligently recommend the appropriate ultrasonic amplitude value. In this way, the influence of too high or too low amplitude on the welding quality can be avoided in the sewing of pull-up pants, thereby improving the welding effect.

[0006] Accordingly, according to one aspect of the present application, an ultrasonic edge sealing process for pull-up pants is provided, which comprises:

[0007] Obtain a side view and a top view of the pull-up pants fabric to be welded;

[0008] The side view of the pull-up pants fabric to be welded is passed through a target detection network to obtain an image of a region of interest;

[0009] Passing the region of interest image through a first convolutional neural network model as a feature extractor to obtain a thickness feature map;

[0010] The top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map;

[0011] fusing the thickness feature map and the material feature map to obtain a decoding feature map; and

[0012] The decoded characteristic graph is passed through a decoder to obtain a decoded value, and the decoded value is used to represent a recommended amplitude value of the ultrasonic wave.

[0013] In the ultrasonic edge sealing process of the above-mentioned pull-up pants, the target detection network is CenterNet, ExtremeNet, RepPoints, YOLOv1 or FCOS.

[0014] In the ultrasonic edge banding process of the above-mentioned pull-up pants, the image of the region of interest is passed through the first convolutional neural network model as a feature extractor to obtain a thickness feature map, including: using each layer of the first convolutional neural network model as a feature extractor to perform convolution processing on the input data in the forward pass of the layer; performing convolution processing on the input data to obtain a convolution feature map; performing pooling on the convolution feature map based on a local feature matrix to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network as a feature extractor is the thickness feature map, and the input of the first layer of the first convolutional neural network as a feature extractor is the image of the region of interest.

[0015] In the ultrasonic edge banding process of the above-mentioned pull-up pants, the top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map, including: extracting a shallow feature map from the Mth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein M is greater than or equal to 1 and less than or equal to 6; extracting a deep feature map from the Nth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein N / M is greater than or equal to 5 and less than or equal to 10; and, using the deep and shallow feature fusion module of the convolutional neural network model including the deep and shallow feature fusion module to fuse the shallow feature map and the deep feature map to obtain the material feature map.

[0016] In the ultrasonic edge-sealing process of the above-mentioned pull-up pants, the thickness characteristic map and the material characteristic map are fused to obtain a decoding characteristic map, including: cascading the thickness characteristic map and the material characteristic map to obtain the decoding characteristic map.

[0017] In the ultrasonic edge sealing process of the pull-up pants, the decoded characteristic graph is passed through a decoder to obtain a decoded value, and the decoded value is used to represent the recommended amplitude value of the ultrasonic wave, including: using the decoder to decode and regress the decoded characteristic graph using the following decoding formula to obtain the decoded value; wherein the decoding formula is: Where X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.

[0018] In the above-mentioned ultrasonic edge banding process of pull-up pants, a training step is also included: training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder; wherein the training step includes: acquiring training data, the training data including a training side view and a training top view of the pull-up pants fabric to be welded, and the true value of the amplitude value of the ultrasonic wave; passing the training side view of the pull-up pants fabric to be welded through the target detection network to obtain a training region of interest image; passing the training region of interest image through the first convolutional neural network model as a feature extractor to obtain a training thickness feature map ; Pass the training top view of the pull-up pants fabric to be welded through the convolutional neural network model including the depth feature fusion module to obtain a training material feature map; fuse the training thickness feature map and the training material feature map to obtain a training decoding feature map; perform Fourier-like scale domain probability correction on the training decoding feature map to obtain an optimized decoding feature map; pass the optimized decoding feature map through a decoder to obtain a decoding loss function value; and, based on the decoding loss function value and through the direction propagation of gradient descent, train the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the depth feature fusion module, and the decoder.

[0019] In the ultrasonic edge sealing process of the pull-up pants, the training decoding feature map is subjected to a Fourier scale domain probability correction to obtain an optimized decoding feature map, including: the training decoding feature map is subjected to a Fourier scale domain probability correction to obtain an optimized decoding feature map according to the following optimization formula; wherein the optimization formula is:

[0020]

[0021] Among them, f i,j,kis the feature value of the (i, j, k)th position of the optimized decoding feature map, f i,j,k is the eigenvalue of the (i, j, k)th position of the decoding feature map, W is the height of the decoding feature map, H is the width of the decoding feature map, C is the number of channels of the decoding feature map, α, β and γ are hyperparameters for scale adjustment, and exp(·) represents the exponential operation of the vector.

[0022] According to another aspect of the present application, an ultrasonic edge sealing system for pull-up pants is also provided, comprising:

[0023] An image extraction module is used to obtain a side view and a top view of the pull-up pants fabric to be welded;

[0024] An image target detection module, used for passing the side view of the pull-up pants fabric to be welded through a target detection network to obtain an image of a region of interest;

[0025] A feature extraction module, used for passing the region of interest image through a first convolutional neural network model as a feature extractor to obtain a thickness feature map;

[0026] A deep and shallow feature fusion module, used for obtaining a material feature map by passing the top view of the pull-up pants fabric to be welded through a convolutional neural network model including the deep and shallow feature fusion module;

[0027] a fusion module, configured to fuse the thickness feature map and the material feature map to obtain a decoding feature map; and

[0028] The decoding module is used to pass the decoding characteristic graph through a decoder to obtain a decoding value, and the decoding value is used to represent the recommended amplitude value of the ultrasonic wave.

[0029] In the ultrasonic edge sealing system of the pull-up pants, the image target detection module is used for: the target detection network is CenterNet, ExtremeNet, RepPoints, YOLOv1 or FCOS

[0030] In the ultrasonic edge sealing system of the above-mentioned pull-up pants, the feature extraction module is used to: use each layer of the first convolutional neural network model as a feature extractor to perform convolution processing on the input data in the forward transfer of the layer; perform convolution processing on the input data to obtain a convolution feature map; perform pooling on the convolution feature map based on the local feature matrix to obtain a pooled feature map; and perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network as a feature extractor is the thickness feature map, and the input of the first layer of the first convolutional neural network as a feature extractor is the region of interest image.

[0031] In the ultrasonic edge sealing system of the above-mentioned pull-up pants, the deep and shallow feature fusion module is used to: extract the shallow feature map from the Mth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein M is greater than or equal to 1 and less than or equal to 6; extract the deep feature map from the Nth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein N / M is greater than or equal to 5 and less than or equal to 10; and, use the deep and shallow feature fusion module of the convolutional neural network model including the deep and shallow feature fusion module to fuse the shallow feature map and the deep feature map to obtain the material feature map.

[0032] In the ultrasonic edge sealing system of the above-mentioned pull-up pants, the fusion module is used to: cascade the thickness feature map and the material feature map to obtain the decoding feature map.

[0033] In the ultrasonic edge sealing system of the pull-up pants, the decoding module is used to: use the decoder to perform decoding regression on the decoding feature map using the following decoding formula to obtain the decoding value; wherein the decoding formula is: Where X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.

[0034] In the ultrasonic edge banding system of the pull-up pants mentioned above, the training step: trains the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder; wherein the training step includes: a training data acquisition unit for acquiring training data, the training data including a training side view and a training top view of the pull-up pants fabric to be welded, and the true value of the amplitude value of the ultrasonic wave; a training area image extraction unit for passing the training side view of the pull-up pants fabric to be welded through the target detection network to obtain a training area of ​​interest image; a training feature map extraction unit for passing the training area of ​​interest image through the first convolutional neural network model as a feature extractor to obtain a training thickness feature map; a training material A material image extraction unit, used to pass the training top view of the pull-up pants fabric to be welded through the convolutional neural network model including the depth feature fusion module to obtain a training material feature map; a fusion unit, used to fuse the training thickness feature map and the training material feature map to obtain a training decoding feature map; a training optimization unit, used to perform Fourier-like scale domain probability correction on the training decoding feature map to obtain an optimized decoding feature map; a classification loss unit, used to pass the optimized decoding feature map through a decoder to obtain a decoding loss function value; and a training unit, used to train the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the depth feature fusion module and the decoder based on the decoding loss function value and through the direction propagation of gradient descent.

[0035] In the ultrasonic edge sealing system of the pull-up pants, the training optimization unit is used to: perform Fourier scale domain probability correction on the training decoding feature map according to the following optimization formula to obtain an optimized decoding feature map; wherein the optimization formula is:

[0036]

[0037] Among them, f i,j,k is the feature value of the (i, j, k)th position of the optimized decoding feature map, f i,j,k is the eigenvalue of the (i, j, k)th position of the decoding feature map, W is the height of the decoding feature map, H is the width of the decoding feature map, C is the number of channels of the decoding feature map, α, β and γ are hyperparameters for scale adjustment, and exp(·) represents the exponential operation of the vector.

[0038] Compared with the prior art, the ultrasonic edge-sealing process and system of pull-up pants provided by the present application adopts artificial intelligence control technology based on deep learning to obtain the side view and top view of the pull-up pants fabric to be welded, and then finally obtains the decoding value through the target detection network, the convolutional neural network of the feature extractor and the decoder, so as to automatically and intelligently recommend the appropriate ultrasonic amplitude value. In this way, the influence of too high or too low amplitude on the welding quality can be avoided in the sewing of pull-up pants, thereby improving the welding effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0040] Figure 1 This is a flow chart of the ultrasonic edge sealing process of pull-up pants according to an embodiment of the present application.

[0041] Figure 2 This is a structural diagram of the ultrasonic edge sealing process of pull-up pants according to an embodiment of the present application.

[0042] Figure 3 The present invention is a flow chart of obtaining a material feature map by passing a top view of the pull-up pants fabric to be welded through a convolutional neural network model including a deep and shallow feature fusion module in the ultrasonic edge sealing process of pull-up pants according to an embodiment of the present application.

[0043] Figure 4 The present invention is a flowchart for training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder in the ultrasonic edge banding process of pull-up pants according to an embodiment of the present application.

[0044] Figure 5 This is a block diagram of an ultrasonic edge sealing system for pull-up pants according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0046] Application Overview

[0047] In response to the above technical problems, the technical concept of the present application is: to obtain the side view and top view of the pull-up pants fabric to be welded to characterize the thickness and material information of the pull-up pants fabric, and to combine deep learning and artificial intelligence technology to automatically and intelligently recommend appropriate ultrasonic amplitude values, thereby avoiding the impact of excessively high or low amplitudes on welding quality.

[0048] Specifically, in the technical solution of the present application, first, a side view and a top view of the pull-up pants fabric to be welded are obtained. Here, by obtaining the side view, the thickness information of the pull-up pants fabric can be obtained; by obtaining the top view, the material characteristics of the pull-up pants fabric, such as texture and color, can be obtained.

[0049] In the actual process of using ultrasound to seal the edges of pull-up pants, the pull-up pants fabric to be welded usually contains multiple parts and layers, and there may be other objects or background interference around it, and the thickness information of the pull-up pants fabric occupies a small proportion in the image, which is easily affected by the interference information. In the technical solution of the present application, the side view of the pull-up pants fabric to be welded is passed through the target detection network to obtain an image of the region of interest, that is, the target detection network is used to automatically detect the area of ​​the pull-up pants fabric in the side view, so as to obtain an image of the region of interest containing only the pull-up pants material. In this way, the interference of the background can be reduced, so as to more accurately extract the thickness characteristics of the pull-up pants material in the subsequent process.

[0050] In order to extract more abstract thickness feature information about the pull-up pants material from the image of the region of interest, in the technical solution of the present application, the image of the region of interest is passed through a first convolutional neural network model as a feature extractor to obtain a thickness feature map. That is, the first convolutional neural network model as a feature extractor is used to automatically extract the thickness features of the pull-up pants material from the image of the region of interest. In this way, the original information in the image of the region of interest can be converted into more representative and separable high-dimensional implicit feature information.

[0051] Considering that the thickness characteristics and material characteristics of the pull-up pants material are both key factors affecting parameter adjustment, in the technical solution of the present application, the top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to extract the texture, color and other material characteristics of the pull-up pants material from the top view, thereby obtaining a material feature map. Here, the convolutional neural network (CNN) in deep learning has strong capabilities in image recognition and processing, and the use of a deep and shallow feature fusion module can fuse features at different levels, retaining the shallow information of the original features, such as the texture, color and other shallow features of the pull-up pants material, further improving the ability to express features, and thus better extracting the material characteristics of the pull-up pants material.

[0052] In order to organically combine the thickness characteristics and material characteristics of the pull-up pants fabric, in the technical solution of the present application, the thickness characteristic map and the material characteristic map are fused to obtain a decoding characteristic map. After obtaining the decoding characteristic map, it is passed through a decoder to obtain a decoding value, and the decoding value is used to represent the recommended amplitude value of the ultrasonic wave. In other words, the decoder is used to convert the thickness characteristic information and material characteristic information contained in the decoding characteristic map into recommended ultrasonic parameters. In practical applications, the subsequent welding process can be guided based on the decoding value, thereby avoiding the influence of too high or too low amplitude on the welding quality, thereby achieving high-quality and high-efficiency welding.

[0053] Here, when the thickness feature map and the material feature map are fused to obtain the decoding feature map, in order to make full use of the thickness characteristics and material characteristics of the pull-up pants fabric to be welded, it is preferred to obtain the decoding feature map by cascading the thickness feature map and the material feature map along the channel dimension. In this way, if the overall feature learning correlation of the first convolutional neural network model serving as a feature extractor and the second convolutional neural network model including the deep and shallow feature fusion module in the three dimensions of their respective width, height and cascaded channels can be improved, the expression effect of the obtained decoding feature map can be further improved.

[0054] Based on this, during the training process, the Fourier-like scale domain probability correction is performed on the decoded feature map, for example, denoted as F, which is specifically expressed as:

[0055]

[0056] Among them, f i，j,k ∈F is the eigenvalue of the (i, j, k)th position of the decoded feature map F, W, H and C are the height, width and number of channels of the decoded feature map F, respectively, and α, β and γ are hyperparameters for scale adjustment.

[0057] Here, the Fourier-like scale domain probability correction takes into account the homology between the high-dimensional feature distribution and the scale domain in which it is located, and can capture the potential distribution association in the homologous space based on the low-rank constraint of the scale space through the Fourier-like sparse low-rank transformation of the scale space. Therefore, in the training process of the first convolutional neural network model as a feature extractor and the second convolutional neural network model including the deep and shallow feature fusion module, while using the convolution kernel to obtain the feature representation of the image semantics and distributing it in the channel dimension, a joint feature learning with the overall scale coherence of the feature map is realized, so as to improve the expression effect of the obtained decoding feature map by improving the learning correlation between the first convolutional neural network model as a feature extractor and the second convolutional neural network model including the deep and shallow feature fusion module at the overall feature map expression scale, thereby improving the accuracy of the decoding value obtained by the decoder of the decoding feature map.

[0058] Based on this, the present application provides an ultrasonic edge banding process for pull-up pants, which includes: obtaining a side view and a top view of the pull-up pants fabric to be welded; passing the side view of the pull-up pants fabric to be welded through a target detection network to obtain an image of a region of interest; passing the image of the region of interest through a first convolutional neural network model as a feature extractor to obtain a thickness feature map; passing the top view of the pull-up pants fabric to be welded through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map; fusing the thickness feature map and the material feature map to obtain a decoded feature map; and, passing the decoded feature map through a decoder to obtain a decoded value, which is used to represent the recommended amplitude value of the ultrasonic wave.

[0059] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0060] Exemplary Methods

[0061] Figure 1 FIG. 1 is a flow chart of an ultrasonic edge sealing process for pull-up pants according to an embodiment of the present application. Figure 1As shown, the ultrasonic edge banding process of pull-up pants according to the embodiment of the present application includes: S110, obtaining a side view and a top view of the pull-up pants fabric to be welded; S120, passing the side view of the pull-up pants fabric to be welded through a target detection network to obtain an image of the region of interest; S130, passing the image of the region of interest through a first convolutional neural network model as a feature extractor to obtain a thickness feature map; S140, passing the top view of the pull-up pants fabric to be welded through a convolutional neural network model including a depth feature fusion module to obtain a material feature map; S150, fusing the thickness feature map and the material feature map to obtain a decoding feature map; and, S160, passing the decoded feature map through a decoder to obtain a decoding value, which is used to represent the recommended amplitude value of the ultrasonic wave.

[0062] Figure 2 FIG. 1 is a schematic diagram of the ultrasonic edge sealing process of pull-on pants according to an embodiment of the present application. Figure 2 As shown, in this architecture, first, a side view and a top view of the pull-up pants fabric to be welded are obtained. Then, the side view of the pull-up pants fabric to be welded is passed through a target detection network to obtain an image of the region of interest. Then, the image of the region of interest is passed through a first convolutional neural network model as a feature extractor to obtain a thickness feature map. At the same time, the top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a depth feature fusion module to obtain a material feature map. Then, the thickness feature map and the material feature map are fused to obtain a decoded feature map. Finally, the decoded feature map is passed through a decoder to obtain a decoded value, which is used to represent the recommended amplitude value of the ultrasonic wave.

[0063] In step S110, a side view and a top view of the pull-up pants fabric to be welded are obtained. Here, by obtaining the side view, the thickness information of the pull-up pants fabric can be obtained; by obtaining the top view, the material characteristics of the pull-up pants fabric, such as texture and color, can be obtained.

[0064] In step S120, the side view of the pull-up pants fabric to be welded is passed through a target detection network to obtain an image of the region of interest. Considering that in the actual process of using ultrasound to seal the pull-up pants, the pull-up pants fabric to be welded usually contains multiple parts and layers, and there may be other objects or background interference around it, and the thickness information of the pull-up pants fabric occupies a small proportion in the image and is easily affected by interference information, in the technical solution of the present application, the side view of the pull-up pants fabric to be welded is passed through a target detection network to obtain an image of the region of interest, that is, the target detection network is used to automatically detect the area of ​​the pull-up pants fabric in the side view, so as to obtain an image of the region of interest containing only the pull-up pants material. In this way, the interference of the background can be reduced, so that the thickness characteristics of the pull-up pants material can be extracted more accurately in the subsequent process.

[0065] Specifically, in an embodiment of the present application, the target detection network is CenterNet, ExtremeNet, RepPoints, YOLOv1 or FCOS.

[0066] In step S130, the image of the region of interest is passed through the first convolutional neural network model as a feature extractor to obtain a thickness feature map. In order to extract more abstract thickness feature information about the pull-up pants material from the image of the region of interest, in the technical solution of the present application, the image of the region of interest is passed through the first convolutional neural network model as a feature extractor to obtain a thickness feature map, that is, the first convolutional neural network model as a feature extractor is used to automatically extract the thickness features of the pull-up pants material from the image of the region of interest, so that the original information in the image of the region of interest can be converted into more representative and separable high-dimensional implicit feature information.

[0067] Specifically, in an embodiment of the present application, the image of the region of interest is passed through a first convolutional neural network model as a feature extractor to obtain a thickness feature map, including: using each layer of the first convolutional neural network model as a feature extractor to perform convolution processing on the input data in the forward pass of the layer; performing convolution processing on the input data to obtain a convolution feature map; performing pooling on the convolution feature map based on a local feature matrix to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network as a feature extractor is the thickness feature map, and the input of the first layer of the first convolutional neural network as a feature extractor is the image of the region of interest.

[0068] In step S140, the top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map. Considering that the thickness characteristics and material characteristics of the pull-up pants material are key factors affecting parameter adjustment, in the technical solution of the present application, the top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to extract the material characteristics of the pull-up pants material, such as texture and color, from the top view, thereby obtaining a material feature map. Here, the convolutional neural network (CNN) in deep learning has a strong ability in image recognition and processing, and the use of a deep and shallow feature fusion module can fuse features at different levels, retaining the shallow information of the original features, such as shallow features such as texture and color of the pull-up pants material, further improving the ability to express features, and thus better extracting the material characteristics of the pull-up pants material.

[0069] Figure 3The present invention is a flowchart of obtaining a material feature map by passing the top view of the pull-up pants fabric to be welded through a convolutional neural network model including a deep and shallow feature fusion module in the ultrasonic edge sealing process of the pull-up pants according to an embodiment of the present application. Figure 3 As shown, the top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map, including: S210, extracting a shallow feature map from the Mth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein M is greater than or equal to 1 and less than or equal to 6; S220, extracting a deep feature map from the Nth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein N / M is greater than or equal to 5 and less than or equal to 10; and, S230, using the deep and shallow feature fusion module of the convolutional neural network model including the deep and shallow feature fusion module to fuse the shallow feature map and the deep feature map to obtain the material feature map.

[0070] In step S150, the thickness characteristic map and the material characteristic map are fused to obtain a decoded characteristic map. In order to organically combine the thickness characteristics and material characteristics of the pull-up pants fabric, in the technical solution of the present application, the thickness characteristic map and the material characteristic map are fused to obtain a decoded characteristic map. After obtaining the decoded characteristic map, it is passed through a decoder to obtain a decoded value, and the decoded value is used to represent the recommended amplitude value of the ultrasonic wave. In other words, the decoder is used to convert the thickness characteristic information and material characteristic information contained in the decoded characteristic map into recommended ultrasonic parameters. In practical applications, the subsequent welding process can be guided based on the decoded value, thereby avoiding the influence of excessively high or low amplitude on the welding quality, thereby achieving high-quality and efficient welding.

[0071] Specifically, in an embodiment of the present application, fusing the thickness characteristic map and the material characteristic map to obtain a decoding characteristic map includes: cascading the thickness characteristic map and the material characteristic map to obtain the decoding characteristic map.

[0072] In step S160, the decoded characteristic graph is passed through a decoder to obtain a decoded value, and the decoded value is used to represent the recommended amplitude value of the ultrasonic wave.

[0073] Specifically, in an embodiment of the present application, the decoded feature graph is passed through a decoder to obtain a decoded value, and the decoded value is used to represent the recommended amplitude value of the ultrasound wave, including: using the decoder to decode and regress the decoded feature graph using the following decoding formula to obtain the decoded value; wherein the decoding formula is: Where X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.

[0074] It is worth mentioning that those skilled in the art should be aware that before applying a deep neural network model for inference, the deep neural network model must be trained so that the deep neural network can implement specific functional capabilities.

[0075] Specifically, in an embodiment of the present application, a training step is also included: training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder.

[0076] Figure 4 The present invention is a flowchart of training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder in the ultrasonic edge sealing process of pull-up pants according to an embodiment of the present application. Figure 4 As shown, the training step includes: S310, obtaining training data, the training data including a training side view and a training top view of the pull-up pants fabric to be welded, and the true value of the amplitude value of the ultrasonic wave; S320, passing the training side view of the pull-up pants fabric to be welded through the target detection network to obtain a training region of interest image; S330, passing the training region of interest image through the first convolutional neural network model as a feature extractor to obtain a training thickness feature map; S340, passing the training top view of the pull-up pants fabric to be welded through the convolutional neural network model including the depth feature fusion module to obtain training material feature map; S350, fusing the training thickness feature map and the training material feature map to obtain a training decoding feature map; S360, performing a Fourier-like scale-domain probability correction on the training decoding feature map to obtain an optimized decoding feature map; S370, passing the optimized decoding feature map through a decoder to obtain a decoding loss function value; and, S380, training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including a deep and shallow feature fusion module, and the decoder based on the decoding loss function value and propagating in the direction of gradient descent.

[0077] In particular, in the technical solution of the present application, when the thickness feature map and the material feature map are fused to obtain the decoding feature map, in order to make full use of the thickness characteristics and material characteristics of the pull-up pants fabric to be welded, it is preferred to obtain the decoding feature map by cascading the thickness feature map and the material feature map along the channel dimension. In this way, if the overall feature learning correlation of the first convolutional neural network model serving as a feature extractor and the second convolutional neural network model including the deep and shallow feature fusion module in the three dimensions of their respective width, height and cascaded channels can be improved, the expression effect of the obtained decoding feature map can be further improved.

[0078] Specifically, in an embodiment of the present application, the training decoding feature map is subjected to a Fourier scale domain probability correction to obtain an optimized decoding feature map, including: performing a Fourier scale domain probability correction on the training decoding feature map according to the following optimization formula to obtain an optimized decoding feature map; wherein the optimization formula is:

[0079]

[0080] Among them, f i,j,k is the feature value of the (i, j, k)th position of the optimized decoding feature map, f i,j,k is the eigenvalue of the (i, j, k)th position of the decoding feature map, W is the height of the decoding feature map, H is the width of the decoding feature map, C is the number of channels of the decoding feature map, α, β and γ are hyperparameters for scale adjustment, and exp(·) represents the exponential operation of the vector.

[0081] In summary, the ultrasonic edge-sealing process of the pull-up pants according to the embodiment of the present application is explained, which uses artificial intelligence control technology based on deep learning to obtain the side view and top view of the pull-up pants fabric to be welded, and then finally obtains the decoded value through the target detection network, the convolutional neural network of the feature extractor and the decoder, so as to automatically and intelligently recommend the appropriate ultrasonic amplitude value. In this way, the influence of too high or too low amplitude on the welding quality can be avoided in the sewing of the pull-up pants, thereby improving the welding effect.

[0082] Exemplary Systems

[0083] Figure 5 FIG. 1 is a block diagram of an ultrasonic edge sealing system for pull-up pants according to an embodiment of the present application. Figure 5 As shown, the ultrasonic edge sealing system 100 for pull-up pants according to the embodiment of the present application includes: an image extraction module 110, used to obtain a side view and a top view of the pull-up pants fabric to be welded; an image target detection module 120, used to pass the side view of the pull-up pants fabric to be welded through a target detection network to obtain an image of the region of interest; a feature extraction module 130, used to pass the image of the region of interest through a first convolutional neural network model as a feature extractor to obtain a thickness feature map; a depth feature fusion module 140, used to pass the top view of the pull-up pants fabric to be welded through a convolutional neural network model including a depth feature fusion module to obtain a material feature map; a fusion module S150, used to fuse the thickness feature map and the material feature map to obtain a decoded feature map; and a decoding module 160, used to pass the decoded feature map through a decoder to obtain a decoded value, and the decoded value is used to represent the recommended amplitude value of the ultrasonic wave.

[0084] In an embodiment of the present application, the image target detection module includes: the target detection network is CenterNet, ExtremeNet, RepPoints, YOLOv1 or FCOS.

[0085] In an embodiment of the present application, the feature extraction module is used to: use each layer of the first convolutional neural network model as a feature extractor to perform the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolutional feature map; pooling the convolutional feature map based on a local feature matrix to obtain a pooled feature map; and, nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolutional neural network as a feature extractor is the thickness feature map, and the input of the first layer of the first convolutional neural network as a feature extractor is the region of interest image.

[0086] In an embodiment of the present application, the deep and shallow feature fusion module is used to: extract a shallow feature map from the Mth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein M is greater than or equal to 1 and less than or equal to 6; extract a deep feature map from the Nth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein N / M is greater than or equal to 5 and less than or equal to 10; and, use the deep and shallow feature fusion module of the convolutional neural network model including the deep and shallow feature fusion module to fuse the shallow feature map and the deep feature map to obtain the material feature map.

[0087] In an embodiment of the present application, the fusion module is used to: cascade the thickness feature map and the material feature map to obtain the decoding feature map.

[0088] In an embodiment of the present application, the decoding module is used to: use the decoder to perform decoding regression on the decoding feature map using the following decoding formula to obtain the decoding value; wherein the decoding formula is: Where X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.

[0089] In an embodiment of the present application, a training step is also included: training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder; wherein the training step includes: a training data acquisition unit for acquiring training data, wherein the training data includes a training side view and a training top view of the pull-up pants fabric to be welded, and a true value of the amplitude value of the ultrasonic wave; a training area image extraction unit for passing the training side view of the pull-up pants fabric to be welded through the target detection network to obtain a training area of ​​interest image; a training feature map extraction unit for passing the training area of ​​interest image through the first convolutional neural network model as a feature extractor to obtain a training thickness feature map; a training material map extraction unit for An extracting unit is used to pass the training top view of the pull-up pants fabric to be welded through the convolutional neural network model including the depth feature fusion module to obtain a training material feature map; a fusion unit is used to fuse the training thickness feature map and the training material feature map to obtain a training decoding feature map; a training optimization unit is used to perform Fourier scale domain probability correction on the training decoding feature map to obtain an optimized decoding feature map; a classification loss unit is used to pass the optimized decoding feature map through a decoder to obtain a decoding loss function value; and a training unit is used to train the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the depth feature fusion module and the decoder based on the decoding loss function value and through the direction propagation of gradient descent.

[0090] In an embodiment of the present application, the training optimization unit is used to: perform Fourier-like scale domain probability correction on the training decoding feature map according to the following optimization formula to obtain an optimized decoding feature map; wherein the optimization formula is:

[0091]

[0092] Among them, f i,j,k is the feature value of the (i, j, k)th position of the optimized decoding feature map, f i,j,k is the eigenvalue of the (i, j, k)th position of the decoding feature map, W is the height of the decoding feature map, H is the width of the decoding feature map, C is the number of channels of the decoding feature map, α, β and γ are hyperparameters for scale adjustment, and exp(·) represents the exponential operation of the vector.

[0093] In summary, the ultrasonic edge-sealing system of the pull-up pants according to the embodiment of the present application is explained, which obtains the side view and top view of the pull-up pants fabric to be welded by adopting the artificial intelligence control technology based on deep learning, and then finally obtains the decoding value through the target detection network, the convolutional neural network of the feature extractor and the decoder, so as to automatically and intelligently recommend the appropriate ultrasonic amplitude value. In this way, the influence of too high or too low amplitude on the welding quality can be avoided in the sewing of the pull-up pants, thereby improving the welding effect.

Claims

1. An ultrasonic edge sealing process for pull-up pants, characterized in that: include: Obtain a side view and a top view of the pull-up pants fabric to be welded; The side view of the pull-up pants fabric to be welded is passed through a target detection network to obtain an image of a region of interest; Passing the region of interest image through a first convolutional neural network model as a feature extractor to obtain a thickness feature map; The top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map; fusing the thickness feature map and the material feature map to obtain a decoding feature map; as well as Passing the decoded characteristic graph through a decoder to obtain a decoded value, wherein the decoded value is used to represent a recommended amplitude value of the ultrasonic wave; Among them, it also includes a training step: training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder; Wherein, the training step includes: Acquire training data, wherein the training data includes a training side view and a training top view of the pull-up pants fabric to be welded, and a true value of the amplitude value of the ultrasonic wave; Passing the training side view of the pull-up pants fabric to be welded through the object detection network to obtain a training region of interest image; Passing the training region of interest image through the first convolutional neural network model as a feature extractor to obtain a training thickness feature map; The training top view of the pull-up pants fabric to be welded is passed through the convolutional neural network model including the deep and shallow feature fusion module to obtain a training material feature map; fusing the training thickness feature map and the training material feature map to obtain a training decoding feature map; Performing Fourier-like scale-domain probability correction on the training decoding feature graph to obtain an optimized decoding feature graph; Passing the optimized decoding feature map through a decoder to obtain a decoding loss function value; and Based on the decoding loss function value and propagating in the direction of gradient descent, the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder are trained; The Fourier-like scale domain probability correction is performed on the training decoding feature map to obtain an optimized decoding feature map, including: Performing Fourier-like scale-domain probability correction on the training decoding feature map using the following optimization formula to obtain an optimized decoding feature map; The optimization formula is: Among them, f′ i,j,k is the feature value of the (i, j, k)th position of the optimized decoding feature map, f i,j,k is the eigenvalue of the (i, j, k)th position of the decoding feature map, W is the height of the decoding feature map, H is the width of the decoding feature map, C is the number of channels of the decoding feature map, β, β and γ are hyperparameters for scale adjustment, and exp(·) represents the exponential operation of the vector.

2. The ultrasonic edge sealing process for pull-up pants according to claim 1, characterized in that: The target detection network is CenterNet, ExtremeNet, RepPoints, YOLOv1 or FCOS.

3. The ultrasonic edge sealing process for pull-up pants according to claim 2, characterized in that: Passing the region of interest image through a first convolutional neural network model as a feature extractor to obtain a thickness feature map, comprising: Using each layer of the first convolutional neural network model as a feature extractor to perform input data in the forward pass of the layer; Perform convolution on the input data to obtain a convolution feature map; Performing pooling on the convolutional feature map based on a local feature matrix to obtain a pooled feature map; and Performing nonlinear activation on the pooled feature map to obtain an activated feature map; Among them, the output of the last layer of the first convolutional neural network serving as a feature extractor is the thickness feature map, and the input of the first layer of the first convolutional neural network serving as a feature extractor is the region of interest image.

4. The ultrasonic edge sealing process for pull-up pants according to claim 3, characterized in that: The top view of the pull-up pants fabric to be welded is passed through a convolutional neural network model including a deep and shallow feature fusion module to obtain a material feature map, including: Extracting a shallow feature map from the Mth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein M is greater than or equal to 1 and less than or equal to 6; Extracting a deep feature map from the Nth layer of the convolutional neural network model including the deep and shallow feature fusion module, wherein N / M is greater than or equal to 5 and less than or equal to 10; and The shallow feature map and the deep feature map are fused using the deep and shallow feature fusion module of the convolutional neural network model including the deep and shallow feature fusion module to obtain the material feature map.

5. The ultrasonic edge sealing process for pull-up pants according to claim 4, characterized in that: Fusion of the thickness characteristic map and the material characteristic map to obtain a decoding characteristic map comprises: cascading the thickness characteristic map and the material characteristic map to obtain the decoding characteristic map.

6. The ultrasonic edge sealing process for pull-up pants according to claim 5, characterized in that: The decoded characteristic graph is passed through a decoder to obtain a decoded value, and the decoded value is used to represent the recommended amplitude value of the ultrasonic wave, including: The decoder is used to perform decoding regression on the decoding feature map using the following decoding formula to obtain the decoding value; wherein the decoding formula is: Where X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.

7. An ultrasonic edge sealing system for pull-up pants, characterized in that: include: An image extraction module is used to obtain a side view and a top view of the pull-up pants fabric to be welded; An image target detection module, used for passing the side view of the pull-up pants fabric to be welded through a target detection network to obtain an image of a region of interest; A feature extraction module, used for passing the region of interest image through a first convolutional neural network model as a feature extractor to obtain a thickness feature map; A deep and shallow feature fusion module, used for obtaining a material feature map by passing the top view of the pull-up pants fabric to be welded through a convolutional neural network model including the deep and shallow feature fusion module; A fusion module, used for fusing the thickness feature map and the material feature map to obtain a decoding feature map; as well as A decoding module, used for passing the decoding characteristic graph through a decoder to obtain a decoding value, wherein the decoding value is used to represent a recommended amplitude value of the ultrasonic wave; The method further includes: training the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder; The target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder are trained, including: Acquire training data, wherein the training data includes a training side view and a training top view of the pull-up pants fabric to be welded, and a true value of the amplitude value of the ultrasonic wave; Passing the training side view of the pull-up pants fabric to be welded through the object detection network to obtain a training region of interest image; Passing the training region of interest image through the first convolutional neural network model as a feature extractor to obtain a training thickness feature map; The training top view of the pull-up pants fabric to be welded is passed through the convolutional neural network model including the deep and shallow feature fusion module to obtain a training material feature map; fusing the training thickness feature map and the training material feature map to obtain a training decoding feature map; Performing Fourier-like scale-domain probability correction on the training decoding feature graph to obtain an optimized decoding feature graph; Passing the optimized decoding feature map through a decoder to obtain a decoding loss function value; and Based on the decoding loss function value and propagating in the direction of gradient descent, the target detection network, the first convolutional neural network model as a feature extractor, the convolutional neural network model including the deep and shallow feature fusion module, and the decoder are trained; The Fourier-like scale domain probability correction is performed on the training decoding feature map to obtain an optimized decoding feature map, including: Performing Fourier-like scale-domain probability correction on the training decoding feature map using the following optimization formula to obtain an optimized decoding feature map; The optimization formula is: Among them, f′ i,j,k is the feature value of the (i, j, k)th position of the optimized decoding feature map, f i,j,k is the eigenvalue of the (i, j, k)th position of the decoding feature map, W is the height of the decoding feature map, H is the width of the decoding feature map, C is the number of channels of the decoding feature map, α, β and γ are hyperparameters for scale adjustment, and exp(·) represents the exponential operation of the vector.

8. The ultrasonic edge sealing system for pull-up pants according to claim 7, characterized in that: The feature extraction module is used to: Using each layer of the first convolutional neural network model as a feature extractor to perform input data in the forward pass of the layer; Perform convolution on the input data to obtain a convolution feature map; Performing pooling on the convolutional feature map based on a local feature matrix to obtain a pooled feature map; as well as Performing nonlinear activation on the pooled feature map to obtain an activated feature map; Among them, the output of the last layer of the first convolutional neural network serving as a feature extractor is the thickness feature map, and the input of the first layer of the first convolutional neural network serving as a feature extractor is the region of interest image.

Images