Automatic injection and dyeing method for visual identification based on improved YOLOv11 instance segmentation network

Abstract

The invention relates to the technical field of intelligent clothing printing and dyeing and automatic textile processing, in particular to an automatic injection and dyeing method for visual recognition based on an improved YOLOv11 instance segmentation network, which comprises the following steps: S1, shooting finished clothing subjected to injection and dyeing to obtain a training data set; s2, marking a training set, and training and building an improved YOLOv11 network model; s3, samples which are not subjected to injection and dyeing are placed on a material frame, and injection and dyeing colors needed by the robot are manually set according to the injection and dyeing effect; s4, shooting the placed sample clothes; s5, after shooting, performing visual identification on the shot picture by using the trained network model to obtain Mask masks of the clothes and the markers in the picture; and S6, carrying out traditional image processing by using Opencv. According to the invention, the improved neural network model can be utilized to carry out visual identification on the injected and dyed clothes, positioning is provided for a subsequent robot to carry out automatic injection and dyeing, the production efficiency is greatly improved, and the manual operation cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent printing and dyeing of clothing and automated processing of textiles, and in particular to an automated dyeing method based on visual recognition using an improved YOLOv11 instance segmentation network. Background Art

[0002] In recent years, with the continuous development of automation technology and intelligent manufacturing technology, the traditional garment injection dyeing process can no longer meet the increasing production efficiency and quality requirements. In order to improve production efficiency and reduce labor costs, more and more garment manufacturers and dyeing factories have begun to try to introduce robotics and computer vision technology to automate the injection dyeing process. However, although robotics can simulate manual operations, it still faces many technical challenges in the execution process, especially in the accurate recognition and positioning of clothing contours.

[0003] In the traditional dyeing process, especially in the stage of coloring clothes, excessive manual work is required. Almost every workstation needs 6-8 workers to complete it. In addition, the dyeing process is easily affected by factors such as workers' operating skills and fatigue. Therefore, how to use computer vision and artificial intelligence technology to achieve accurate visual recognition of clothing has become one of the core technologies of the automated dyeing system. Due to the irregular shape of clothes in the dyeing process, the penetration of pigments, the gradual change of colors, and the fact that the dyeing workstation is carried out on the material frame, which is a transparent mesh frame, there will be excess pigment penetration. These factors will affect visual recognition. How to achieve high-precision clothing contour recognition and landmark positioning under complex textures and background interference, as well as what appropriate network model should be used for visual recognition, is the key to the successful application of this technology.

[0004] The existing YOLOv11 (You Only Look Once) deep learning model has become one of the mainstream methods in the current visual recognition field due to its high efficiency and accuracy in target detection and instance segmentation tasks. In particular, in the instance segmentation task, YOLOv11 can accurately segment the target and recognize the contours of irregularly shaped objects. However, the existing YOLOv11 instance segmentation model still has certain limitations for complex backgrounds, texture transitions, and color interference. In particular, there is still room for improvement in the accurate recognition of clothing contours and small target markers during the dyeing process.

[0005] Patent CN113684633B proposes an automatic injection dyeing machine and also mentions the relevant dyeing proofing technology. It realizes the positioning and dyeing of clothing through mechanical positioning and fixings. However, injection-dyed fabrics have large shape variability and complex contours, and it is difficult to achieve precise positioning through mechanical methods. Patent CN115710787B proposes a human-machine collaborative intelligent color spraying robot. However, although the equipment has partially realized automation in dyeing operations, its operation process still relies on manual intervention. Operators need to continuously monitor the nozzle blockage status during the production process and manually clean and adjust the equipment operation when necessary.

[0006] In order to overcome the above technical limitations, the present invention automatically adjusts the spraying path and dye spraying strategy through visual recognition, realizes the intelligence and automation of the entire process, and is suitable for mass production of complex textures and personalized patterns. Summary of the invention

[0007] The purpose of the present invention is to solve the shortcomings existing in the prior art, and proposes an automated dyeing method based on visual recognition based on an improved YOLOv11 instance segmentation network. The YOLOv11 instance segmentation network is introduced into the automated dyeing system, and the dye injection and positioning operations are dynamically controlled by real-time identification of clothing contours and marker areas, which significantly improves the automation level and dyeing accuracy of the dyeing process.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] An automated coloring method for visual recognition based on an improved YOLOv11 instance segmentation network comprises the following steps:

[0010] Step 1: Prepare training data set

[0011] Set up the shooting environment and field of view, place the finished dyed clothes on the material frame and shoot them, so as to obtain multiple training data sets;

[0012] Step 2: Build a neural network

[0013] Label the clothing outlines and landmarks in the training data set, train these two features after labeling, and build an improved YOLOv11 neural network model;

[0014] Step 3: Manually set the robot program

[0015] After the model is trained, the undyed clothing samples are placed on the material frame, and the robot's execution program is set manually according to the desired dyeing effect, such as the number and type of colors that the clothing wants to be dyed.

[0016] After the program is set, save it, and the robot can then specify the injection color of different areas according to the program.

[0017] Step 4: Take pictures of sample clothes

[0018] Take photos of the sample clothes that have been placed;

[0019] Step 5: Visual recognition

[0020] After shooting, use the trained network model to perform visual recognition on the captured image to obtain the mask of the clothes and landmarks in the image;

[0021] Step 6: Image processing

[0022] Using Opencv traditional image processing, the recognized images are processed. First, the outer contour point set of the clothes and the marker mask is obtained, and the contour is obtained by connecting them. Then, the marker line is obtained based on the rotation center of the clothes and the center of the marker, connecting the center of the circle, the marker, and then the outer contour of the clothes.

[0023] Step 7: Robotic processing and dyeing

[0024] The robot performs dyeing processing, dyeing within the outline of the clothes, and changing the dyeing color at the marked line position.

[0025] Preferably, in step 1, the specific steps are as follows:

[0026] Step 1.1. The shooting environment should minimize the effect of ambient light changes on subsequent visual recognition, so two strip light sources are added, and their brightness and camera exposure parameters are adjusted to meet the requirements;

[0027] Step 1.2: The captured data set should minimize the impact of the background on subsequent recognition and unify the field of view, so adjust the field of view to just fill the entire frame.

[0028] Among them, the two features of clothes and markers in the training set are aimed at identifying the masks of clothes and markers in the injection dyeing proofing task. The color changes of dyes, the complexity of clothes contours, and the fact that the material frame is a permeable mesh frame, the injected dye can easily penetrate under the material frame to a greater or lesser extent, thus easily affecting the recognition effect, so the recognition accuracy is highly challenging.

[0029] Preferably, in step 2, the specific steps of the improved YOLOv11 neural network model are as follows:

[0030] The CBAM (Convolutional Block Attention Module) module is embedded in the feature fusion stage of YOLOv11. Since the main task of the Neck part in the YOLOv11 network architecture is to fuse feature maps from different resolutions, the introduction of CBAM into the C3k2 module can not only make the model more perceptive to details (such as clothing edges, colors, etc.), more effectively process these difficult-to-distinguish details, reduce the interference of the frame background, but also help the network focus on important features at different scales, thereby improving the recognition accuracy of the large detection head (Head-Large) for clothing contours and the small detection head (Head-Small) for landmarks;

[0031] Among them, the CBAM module is essentially an attention module integrated in the convolutional neural network. Its purpose is to enhance the feature expression ability of the model. The CBAM module consists of a channel attention module (Channel Attention Module) and a spatial attention module (SpatialAttention Module); they are applied to the two dimensions of the feature map: the channel dimension and the spatial dimension, and the features are enhanced in the channel dimension and the spatial dimension.

[0032] Preferably, the channel attention module comprises the following steps:

[0033] a. Perform global average pooling and global max pooling on the feature map to obtain a global descriptor;

[0034] b. Reduce the dimension and activate the descriptor through a fully connected layer (usually ReLU activation);

[0035] c. Finally, the weight of each channel is obtained through sigmoid activation, and these weights are used to weight the channels of the feature map;

[0036] The spatial attention module includes the following steps:

[0037] a. Perform average pooling and maximum pooling on the feature map in the channel dimension to obtain two single-channel feature maps;

[0038] b. Concatenate these two feature maps and pass them through a convolutional layer to obtain a spatial attention weight map;

[0039] c. Finally, the spatial weight is limited to the interval [0,1] through the sigmoid activation function, and each position of the feature map is weighted according to this weight.

[0040] Preferably, after having the spatial attention module and the channel attention module, CBAM mixes the channel attention module and the spatial attention module in series and performs two enhancement processes on the feature map. The specific steps are as follows:

[0041] a. Channel attention enhancement: First, the input feature map F passes through the channel attention module to generate attention weights and perform weighted processing;

[0042] b. Spatial attention enhancement: Then, the feature map after channel enhancement is passed to the spatial attention module, and the spatial attention weight is generated and weighted processing is performed to obtain the final output feature map.

[0043] The CBAM module is introduced here. That is, after the introduction of these two attention mechanisms, through channel attention, CBAM can help YOLOv11 pay more attention to the channels of important features such as the outlines of clothes and markers and the boundaries of dyes in complex scenes, avoiding interference from background or irrelevant areas. Through spatial attention, CBAM can dynamically adjust attention at different spatial positions. Because the clothes to be dyed are all on the material frame, the spatial attention module will enhance the features of important areas while suppressing the influence of irrelevant areas, that is, ensuring that the model can pay attention to the key areas within the material frame, thereby improving segmentation accuracy and reducing interference from irrelevant information. It is worth mentioning that the CBAM module is relatively lightweight and will not significantly increase the computational overhead compared to other attention mechanism modules. It is suitable for use in combination with YOLOv11's efficient detection and segmentation architecture.

[0044] Subsequently, the improved YOLOv11 instance segmentation network model is trained, and the training results are exported as .onnx files and deployed into the visual recognition program.

[0045] Preferably, step 6 includes the following steps:

[0046] a. First obtain the outer contour point set of the clothes and the marker mask, and connect them according to the outer contour point set to obtain the outer contour of the clothes;

[0047] b. Get the center of the clothing contour O1 and the center of other marker contours O2, O3…O n ;

[0048] c. Starting from O1, follow the path from the center point O i (i=2,3…n) and find the intersection point O with the clothes outline i ', connect O1 O i ' is the marking line.

[0049] Preferably, step 7 includes the following steps:

[0050] a. After the robot obtains the position coordinate information of the clothes outline, the program sets it to process and color the inside of the clothes outline;

[0051] b. According to the number and position information of the markers, the program sets the processing area, and different processing areas are dyed with different colors according to the manually set colors.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present invention is used to accurately identify the outline and markers of clothing during the dyeing proofing stage. By improving the YOLOv11 model, it can still maintain high-precision recognition performance when facing challenges such as complex dyeing textures, color gradients, and pigment penetration. This improvement can not only smoothly identify the outlines and markers of different clothes, but also provide precise positioning and control for the robot in the subsequent automated dyeing process, ultimately achieving full automation of the dyeing task, greatly improving production efficiency, reducing manual operation costs, and improving the quality control capability of the dyeing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of a sample garment before dyeing in the present invention;

[0055] Figure 2 It is a schematic diagram of the details of the material frame in the present invention;

[0056] Figure 3 This is the improved network structure diagram of YOLOv11 in the present invention;

[0057] Figure 4 Schematic diagram of the principle structure of the channel attention module in the present invention;

[0058] Figure 5 Schematic diagram of the principle structure of the spatial attention module in the present invention;

[0059] Figure 6 It is a principle structure diagram of the CBAM hybrid attention module in the present invention;

[0060] Figure 7 This is a schematic diagram of the processing result of using Opencv to process the recognized image to obtain the outer contour of the clothes and the marking line in the present invention;

[0061] Figure 8 This is a schematic diagram of the processing effect after robot dyeing in the present invention;

[0062] Fig. 9 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the protection scope of the present invention. The embodiments described in the present invention are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0064] An automated coloring method for visual recognition based on an improved YOLOv11 instance segmentation network comprises the following steps:

[0065] (1) Prepare a training data set;

[0066] First, set up the shooting environment and adjust the camera's field of view. The shooting environment should minimize the effect of ambient light changes on subsequent visual recognition, so an additional light source is added and its brightness and camera exposure parameters are adjusted to meet the requirements. The captured data set should minimize the impact of the background on subsequent recognition and unify the field of view, so the field of view is adjusted to just fill the entire frame.

[0067] Then put the undyed sample clothes on the material frame, and post the markers for the color change positions and take pictures. Figure 1 That is, for the circular sample clothes that have not been dyed, the various postures of the clothes should be changed as much as possible during the shooting of the data set to improve the robustness of recognition, so as to obtain multiple training data sets.

[0068] (2) Constructing a neural network;

[0069] After shooting, the clothing outlines and markers in the training set images are marked. After marking, the two features can be sorted out for the training set and verification set required for training. In the injection dyeing proofing task, the goal is to identify the outline of the clothing and the rectangular color markers. The color changes of the dyes, the complexity of the clothing outlines, and the fact that the material frame is a permeable mesh frame, such as Figure 2 As shown in the detailed picture of the screen frame, the injected pigment can easily penetrate into the bottom of the frame to a greater or lesser extent, thus affecting the recognition effect, so the recognition accuracy is highly challenging.

[0070] YOLOv11 is the latest generation of the YOLO (You Only LookOnce) series of neural networks, which further improves the performance of object detection and instance segmentation tasks. As a deep learning network with strong real-time performance, YOLOv11 combines efficient feature extraction capabilities and a multi-task learning framework, and achieves higher detection and segmentation accuracy through improved network architecture and loss function.

[0071] YOLOv11's instance segmentation capability is based on the pixel-level segmentation feature introduced in the target detection framework. Its design allows accurate segmentation of each pixel of the target while detecting the target bounding box. This capability enables YOLOv11 to accurately predict the boundary of target shapes in complex scenes while maintaining efficiency and robustness in segmentation tasks.

[0072] Since the segmentation of clothing contours is relatively complex, the use of YOLOv11 will still result in inaccurate segmentation results. Therefore, in order to further improve its performance, certain improvements have been made to the basic network architecture of YOLOv11, and the CBAM (Convolutional Block Attention Module) module has been embedded in the feature fusion stage of YOLOv11. Since the main task of the Neck part in the YOLOv11 network architecture is to fuse feature maps from different resolutions, the introduction of CBAM into the C3k2 module of the Neck part can not only make the model more perceptive of details (such as clothing edges, colors, etc.), but also more effectively process these difficult-to-distinguish details, reduce the interference of the frame background, and help the network focus on important features at different scales, thereby improving the recognition accuracy of the large detection head (Head-Large) for clothing contours and the small detection head (Head-Small) for landmarks. The improved network architecture is as follows: Figure 3 shown.

[0073] The CBAM module is essentially an attention module integrated into the convolutional neural network. Its purpose is to enhance the model's feature expression ability. It consists of a channel attention module (ChannelAttentionModule) and a spatial attention module (SpatialAttentionModule). They can be applied to the two dimensions of the feature map: the channel dimension and the spatial dimension, and enhance the features in the channel dimension and the spatial dimension. The following is an analysis of these two modules in CBAM:

[0074] ① Channel attention mechanism:

[0075] like Figure 4 As shown in the schematic diagram of the channel attention module, the processing steps are as follows:

[0076] a. For the input feature map F∈R C×H×W (where C is the number of channels, H and W represent the height and width respectively), first perform global maximum pooling and global average pooling operations on each channel to obtain the global description F avg ∈R C×1×1 and F max ∈R C×1×1 :

[0077]

[0078] In the formula, F avg (c) represents the average pooling result of the cth channel, F max (c) shows the result of average pooling of the c-th channel.

[0079] b. After extracting the feature vector through global maximum pooling and global average pooling operations, these two types of features are passed into a shared fully connected layer. This fully connected layer is used to learn the attention weights of each channel, so that the network can adaptively determine which channels are more important for the current task. Finally, the final attention weight vector is generated by combining the feature vectors of global maximum pooling and average pooling.

[0080] MLP(x) = W1·ReLU(W0·x)

[0081] Where W0 and W1 are learnable weight matrices, and ReLU is a nonlinear activation function.

[0082] c. To limit the attention weights to the range of 0 to 1, a Sigmoid activation function is applied to generate channel attention weights. These weights are then applied to each channel of the original feature map.

[0083] M c =σ(MLP(F avg )+MLP(F max ))

[0084] Where σ is the sigmoid function, which ensures that the output is between 0 and 1.

[0085] d. Multiply the generated attention weights by each channel of the original feature map to obtain a weighted channel feature map. This operation can highlight the channels that contribute to the current task while suppressing channels that are not related to the task.

[0086] F′=M c ⊙F

[0087] Where ☉ represents element-wise multiplication.

[0088] ②Spatial Attention Mechanism:

[0089] like Figure 5 As shown in the schematic diagram of the spatial attention module, the processing steps are as follows:

[0090] a. For the input feature map F∈R C×H×W(where C is the number of channels, H and W represent the height and width respectively), first perform global maximum pooling and global average pooling operations on each channel to obtain a global description and

[0091]

[0092] In the formula represents the channel average at position (i, j), Indicates the maximum value of the channel at position (i, j).

[0093] b. The features obtained by maximum pooling and average pooling are concatenated along the channel dimension to form a feature map that integrates contextual information of different scales.

[0094]

[0095] Where [;] represents the concatenation operation in the channel dimension.

[0096] c. Next, the feature map is convolved with a 7×7 convolution kernel to generate the corresponding spatial attention weights. The sigmoid activation function is then used to limit the spatial weights to the interval [0, 1].

[0097]

[0098] Where σ is the sigmoid function, f 7×7 It is a 7×7 convolution operation.

[0099] d. Apply the generated spatial attention weights to the original feature map to weight the features at each spatial position, thereby highlighting the key image areas while reducing the influence of unimportant areas.

[0100] F″=M S ⊙F

[0101] After having the spatial attention module and the channel attention module, CBAM combines the channel attention module and the spatial attention module in series and performs two enhancements on the feature map. Figure 6 As shown, the whole process is as follows:

[0102] a. First, the input feature map F passes through the channel attention module to generate the attention weight M c And perform weighted processing to obtain the enhanced feature map F c .

[0103] b. Then, the feature map F after channel enhancement c It is passed into the spatial attention module to generate the spatial attention weight Ms And perform weighted processing to obtain the final output feature map F final , the processing flow of the entire CBAM module can be expressed as:

[0104] F final =M s (M c (F)⊙F)⊙F

[0105] After the introduction of the CBAM module, that is, the introduction of these two attention mechanisms, through channel attention, CBAM can help YOLOv11 pay more attention to the channels of important features such as the outlines of clothes and markers and the boundaries of dyes in complex scenes, avoiding interference from background or irrelevant areas. Through spatial attention, CBAM can dynamically adjust attention at different spatial positions. Because the clothes that need to be dyed are all on the material frame, the spatial attention module will enhance the features of important areas while suppressing the influence of irrelevant areas, that is, ensuring that the model can pay attention to the key areas within the material frame, thereby improving segmentation accuracy and reducing interference from irrelevant information. It is worth mentioning that the CBAM module is relatively lightweight and will not significantly increase the computational overhead compared to other attention mechanism modules. It is suitable for use in combination with YOLOv11's efficient detection and segmentation architecture.

[0106] Subsequently, the improved YOLOv11 instance segmentation network model is trained, and the training results are exported as .onnx files and deployed into the visual recognition program.

[0107] (3) Manually set the robot program;

[0108] After the model is trained, the undyed clothing samples are placed on the material frame, and the robot's related execution programs are manually set according to the desired dyeing effect, such as the number and type of colors that the clothes want to be dyed.

[0109] After the program is set, save it, and the robot can then specify the injection color of different areas according to the program.

[0110] (4) Take pictures of sample clothing;

[0111] Take photos of the sample clothes that have been placed.

[0112] (5) Visual recognition;

[0113] The improved YOLOv11 instance segmentation network model is used to recognize the pictures taken after teaching, and the masks of the clothes outline and the landmark outline are finally recognized.

[0114] (6) Image processing

[0115] Use Opencv traditional image processing to process the recognized images:

[0116] a. First obtain the outer contour point set of the clothes and the marker mask, and connect them according to the outer contour point set to obtain the outer contour of the clothes.

[0117] b. Get the center of the clothing contour O1 and the center of other marker contours O2, O3... n .

[0118] c. Starting from O1, follow the path from the center point O i (i=2,3…n) and find the intersection point O with the clothes outline i ', connect O1O i ′ is the marking line.

[0119] The outer contour and marking lines of the processed clothes are as follows Figure 7 shown.

[0120] (7) Robotic processing and dyeing

[0121] The robot performs dyeing and injection processing, injects dyeing within the outline of the clothes, and changes the dyeing color at the marking line position. The effect after injection dyeing is as follows Figure 8 As shown, the robot replaces manual labor to automate the injection and dyeing process, achieving the same expected effect and reducing manual operation costs.

[0122] In summary, the present invention is used to accurately identify the contours and markers of clothing during the dyeing proofing stage. By improving the YOLOv11 model, it can still maintain high-precision recognition performance when facing challenges such as complex dyeing textures, color gradients, and pigment penetration. This improvement can not only smoothly identify the contours and markers of different clothes, but also provide precise positioning and control for the robot in the subsequent automated dyeing process, ultimately achieving full automation of the dyeing task, greatly improving production efficiency, reducing manual operation costs, and improving the quality control capabilities of the dyeing process.

[0123] The description and practice disclosed in the present invention are easy to think and understand for ordinary technicians in the technical field, and several improvements and modifications can be made without departing from the principles of the present invention. Therefore, modifications or improvements made without departing from the spirit of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An automated coloring method for visual recognition based on an improved YOLOv11 instance segmentation network, characterized in that: The steps include: Step 1: Prepare training data set Set up the shooting environment and field of view, place the finished dyed clothes on the material frame and shoot them, so as to obtain multiple training data sets; Step 2: Build a neural network Label the clothing outlines and landmarks in the training data set, train these two features after labeling, and build an improved YOLOv11 neural network model; Step 3: Manually set the robot program Place the undyed sample on the material frame, and manually set the dyeing color required by the robot according to the dyeing effect; Step 4: Take pictures of sample clothes Take photos of the sample clothes that have been placed; Step 5: Visual recognition After shooting, use the trained network model to perform visual recognition on the captured image to obtain the mask of the clothes and landmarks in the image; Step 6: Image processing Using Opencv traditional image processing, the recognized images are processed to obtain the location of the clothing outline and logo line; Step 7: Robotic processing and dyeing The robot performs dyeing processing, dyeing within the outline of the clothes, and changing the dyeing color at the marked line position.

2. The method for automatic coloring based on improved YOLOv11 instance segmentation network for visual recognition according to claim 1, characterized in that: In step 1, the specific steps are as follows: Step 1.1: The shooting environment should reduce the effect of ambient light changes on subsequent visual recognition. Add two bar light sources and adjust their brightness and camera exposure parameters to meet the requirements. Step 1.2: The captured data set should avoid the influence of the background on subsequent recognition and unify the field of view, and adjust the field of view to fill the entire frame.

3. The method for automatic coloring based on improved YOLOv11 instance segmentation network for visual recognition according to claim 1, characterized in that: In step 2, the specific steps of the improved YOLOv11 neural network model are as follows: The CBAM module is embedded in the feature fusion stage of YOLOv11. The CBAM module is an attention module integrated in the convolutional neural network. The purpose is to enhance the feature expression ability of the model. The CBAM module consists of a channel attention module and a spatial attention module. It is applied to the two dimensions of the feature map: the channel dimension and the spatial dimension, and the features are enhanced in the channel dimension and the spatial dimension.

4. The method for automatic coloring based on improved YOLOv11 instance segmentation network for visual recognition according to claim 3, characterized in that: The channel attention module includes the following steps: a. Perform global average pooling and global maximum pooling on the feature map to obtain a global descriptor; b. Reduce the dimension and activate the descriptor through a fully connected layer; c. Finally, the weight of each channel is obtained through sigmoid activation, and these weights are used to weight the channels of the feature map; The spatial attention module includes the following steps: a. Perform average pooling and maximum pooling on the feature map in the channel dimension to obtain two single-channel feature maps; b. Concatenate these two feature maps and pass them through a convolutional layer to obtain a spatial attention weight map; c. Finally, the spatial weight is limited to the interval [0,1] through the sigmoid activation function, and each position of the feature map is weighted according to this weight.

5. The method for automatic coloring based on improved YOLOv11 instance segmentation network for visual recognition according to claim 3, characterized in that: By introducing the channel attention module and the spatial attention module, the specific steps for enhancing features in the channel dimension and the spatial dimension are as follows: a. Channel attention enhancement: First, the input feature map F passes through the channel attention module to generate attention weights and perform weighted processing; b. Spatial attention enhancement: Then, the feature map after channel enhancement is passed to the spatial attention module, and the spatial attention weight is generated and weighted processing is performed to obtain the final output feature map.

6. The method for automatic coloring based on improved YOLOv11 instance segmentation network for visual recognition according to claim 1, characterized in that: Step 6 includes the following steps: a. First obtain the outer contour point set of the clothes and the marker mask, and connect them according to the outer contour point set to obtain the outer contour of the clothes; b. Get the center of the clothing contour O1 and the center of other marker contours O2, O3…O n ; c. Starting from O1, follow the path from the center point O i (i=2,3…n) and find the intersection point O with the clothes outline i ', connect to O1O i ' is the marking line.

7. The method for automatic coloring based on improved YOLOv11 instance segmentation network for visual recognition according to claim 1, characterized in that: Step 7 includes the following steps: a. After the robot obtains the position coordinate information of the clothes outline, the program sets it to process and color the inside of the clothes outline; b. According to the number and position information of the markers, the program sets the processing area, and different processing areas are dyed with different colors according to the manually set colors.

Images