An intelligent inkjet control system, method, robot and storage medium

Abstract

This invention relates to the field of inkjet control technology, and discloses an intelligent inkjet control system, method, robot, and storage medium. The method first acquires the surface contour data and color information of the object to be inkjetted using a laser rangefinder, constructing a surface feature vector. Then, it determines the target inkjet area through a region suitability assessment. Next, it acquires the semantic information of the inkjet content, calculates the semantic element weights, and generates inkjet parameter constraints, including droplet size thresholds and information hierarchy presentation rules, based on the target area range. Subsequently, it generates an intelligent inkjet scheme based on the constraints, controlling the inkjet device to form the target pattern. Finally, it generates the movement path of a mobile robotic arm based on a cubic spline interpolation algorithm, applying the target pattern to the target area. This method can adapt to complex surface morphologies and diverse materials, precisely control the inkjet process, ensure accurate pattern presentation and stable adhesion, and improve inkjet effect and information transmission effectiveness.

Description

Technical Field

[0001] This invention relates to the field of inkjet control technology, specifically to an intelligent inkjet control system, method, robot, and storage medium. Background Technology

[0002] In practical applications of inkjet printing technology, the surface morphology of the objects to be inkjet printed often varies significantly, including planar structures as well as complex three-dimensional structures such as curved surfaces and uneven surfaces. Traditional inkjet control methods are mostly based on preset planar printing parameters, which are difficult to adapt to the different surface morphological characteristics of objects. This leads to problems such as deformation, blurring, or even detachment of the inkjet pattern in non-planar areas. For example, when inkjet printing on curved objects, if the droplet size and jet angle of planar printing are still used, the ink droplet distribution is easily uneven due to changes in surface curvature, affecting the integrity and clarity of the pattern.

[0003] Different materials exhibit significantly different ink adhesion properties. Materials such as metals, plastics, and ceramics have varying surface smoothness and porosity. Traditional methods, lacking specific inkjet parameters tailored to these material characteristics, often result in poor ink adhesion, cracking after drying, or wiping off. Furthermore, the adaptation of inkjet content to the surface area of ​​the object is insufficient. When inkjet content contains multiple layers of semantic information, traditional control methods cannot rationally allocate information hierarchy according to the area range, leading to the compression of important information or the over-presentation of secondary information, thus affecting the effectiveness of information transmission.

[0004] Current inkjet systems often rely on manual judgment or simple region division rules to determine the target inkjet area, lacking accurate assessment of region suitability for inkjet printing. Some areas may be unsuitable for inkjet printing due to material properties or surface conditions, but traditional methods cannot identify such areas. Forcing inkjet printing into these areas not only wastes ink but may also damage the object's surface. Furthermore, the path planning of mobile robotic arms often uses linear interpolation algorithms, which are prone to path jamming and large positioning errors when dealing with complex curved surfaces or large inkjet areas. This results in the inkjet pattern failing to accurately conform to the target area, reducing the consistency and stability of the inkjet printing effect.

[0005] With the increasing demand for personalized inkjet printing in industrial production, art processing, and other fields, the semantic information presentation of inkjet content is becoming increasingly important. Traditional methods struggle to differentiate the presentation of inkjet content containing multiple layers of semantics based on the importance of semantic elements, leading to the weakening of key information and affecting the accuracy of information delivery. For example, in product labeling inkjet printing, the brand logo and explanatory text have different levels of importance, but traditional methods often use the same printing parameters, failing to highlight the core information. Furthermore, the lack of correlation between parameters such as droplet size and pattern size and semantic information results in inconsistent presentation of inkjet content in different areas, making it difficult to meet diverse application needs.

[0006] In existing inkjet control methods, the coordination between the inkjet unit and the mobile robotic arm is not close enough. The position information of both is not effectively integrated into the path planning, which makes the robotic arm prone to positioning deviations during movement, affecting the final position of the inkjet pattern. Especially when handling large objects or multi-area inkjet tasks, unreasonable path planning increases the robotic arm's movement time, reduces overall inkjet efficiency, and makes it difficult to meet the requirements of high-efficiency production. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent inkjet control system, method, robot, and storage medium to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides an intelligent inkjet control method, the method comprising:

[0009] The surface contour data and color information of the object to be inkjet are obtained by a preset laser rangefinder and used to construct an object surface feature vector that includes surface morphology features and material properties.

[0010] The surface feature vector of the object is evaluated for regional inkjetability to obtain an inkjetable region distribution map including the probability distribution of inkjetable regions, and the target inkjet region is determined according to the highest confidence value corresponding to the inkjetable region distribution map.

[0011] Obtain the semantic information corresponding to the preset inkjet content, calculate the semantic element weights corresponding to the semantic information, and generate inkjet parameter constraints corresponding to the inkjet content based on the area range, semantic element weights, and semantic information corresponding to the target inkjet area; the inkjet parameter constraints include at least droplet size thresholds, information hierarchy presentation rules, inkjet pattern size parameters, and material adhesion firmness parameters.

[0012] Based on the inkjet parameter constraints, an intelligent inkjet scheme corresponding to the inkjet content is generated, and the preset inkjet device is controlled to perform inkjet printing according to the intelligent inkjet scheme to form the target inkjet pattern.

[0013] Based on the cubic spline interpolation algorithm, a preset movement path corresponding to the mobile robotic arm is generated according to the first position information of the object to be inkjetized, the second position information of the preset inkjet device, and the area range corresponding to the target area. The mobile robotic arm is then controlled to apply the target inkjet pattern to the target area according to the movement path.

[0014] Preferably, the step of acquiring the surface contour data and color information of the object to be inkjetized using a preset laser ranging device, and constructing an object surface feature vector containing surface morphology features and material properties, includes:

[0015] The color information is input into a preset recurrent neural network to extract color feature maps of multiple color channels;

[0016] The surface contour data is input into a preset point cloud processing network to perform surface morphology modeling on the contour data and output surface morphology features including spectral reflectance and material properties.

[0017] The color feature map and the surface morphology feature are fused into feature vectors. The fused feature vectors are then compressed in dimension through normalization to generate object surface feature vectors with local morphology preservation.

[0018] Preferably, the step of evaluating the inkjetability of the object surface feature vector to obtain an inkjetable region distribution map including the probability distribution of inkjetable regions includes:

[0019] A deep region segmentation network is constructed. The deep region segmentation network adopts the ResNet architecture for encoding and decoding. In the encoding stage, a spatial attention module is introduced to weight the importance of the feature vectors on the object surface. In the decoding stage, a dilated convolutional layer is embedded to dynamically adjust the receptive field to adapt to the contour features of different shaped regions.

[0020] The surface feature vector of the object is input into a depth region segmentation network, and the output is a spraying probability map containing the spraying suitability of each pixel.

[0021] The appropriate spray probability map is processed by median filtering to eliminate discrete noise and generate an appropriate spray region distribution map with regional connectivity.

[0022] Preferably, determining the target inkjet region based on the highest confidence value corresponding to the suitable inkjet region distribution map includes:

[0023] An evaluation index system is constructed based on the maximum coverage area, minimum surface roughness variance, and maximum edge sharpness index corresponding to the target inkjet area.

[0024] The suitable inkjet region distribution map is optimized according to the preset evaluation index system and particle swarm optimization algorithm to obtain multiple candidate inkjet regions;

[0025] The MOEA / D algorithm is used to perform non-dominated sorting of multiple candidate inkjet regions, and the comprehensive score corresponding to each candidate inkjet region after sorting is calculated according to the analytic hierarchy process.

[0026] Among multiple candidate inkjet regions, the continuous region with a comprehensive score greater than the threshold value is selected as the target inkjet region.

[0027] Preferably, the step of calculating the semantic element weights corresponding to the semantic information, and generating inkjet parameter constraints corresponding to the inkjet content based on the region range corresponding to the target inkjet region, the semantic element weights, and the semantic information, includes:

[0028] Extract the semantic association graph corresponding to the semantic information based on the preset LSTM model;

[0029] The importance score of each semantic node in the semantic association graph is calculated based on a preset graph neural network and used as the weight of the semantic element.

[0030] The region range parameters corresponding to the region range are transformed into nozzle arrangement constraints, and the semantic element weights are mapped into weighted coefficients of information hierarchy presentation rules; the region range parameters include one or more of the following: geometric size parameters, morphological feature parameters, edge feature parameters, material property parameters, spatial position and orientation parameters, topological parameters, and stability parameters;

[0031] The inkjet parameter constraints corresponding to the inkjet content are obtained based on the printhead arrangement constraints and weighting coefficients using a dynamic programming method.

[0032] Preferably, the step of generating the intelligent inkjet solution corresponding to the inkjet content based on the inkjet parameter constraints includes:

[0033] A smart inkjet architecture based on a reinforcement learning policy network is constructed. The policy network of the reinforcement learning policy network adopts a multi-scale feature fusion network with a multi-head attention mechanism, and the value network of the reinforcement learning policy network introduces a simulation rendering module to simulate the actual inkjet effect. The policy network parameters of the reinforcement learning policy network are optimized based on the policy gradient, and the corresponding reward function of the reinforcement learning is generated according to information entropy, visual coordination degree and material matching degree.

[0034] A style parameter vector is generated based on preset visual specifications and inkjet parameter constraints, and the style parameter vector is input into the feature space corresponding to the intelligent inkjet architecture.

[0035] The inkjet content is input into the intelligent inkjet architecture to generate the intelligent inkjet solution.

[0036] Preferably, based on a cubic spline interpolation algorithm, a preset movement path for the mobile robotic arm is generated according to the first position information corresponding to the object to be inkjetized, the preset second position information of the inkjet device, and the area range corresponding to the target area, including:

[0037] A kinematic model based on Cartesian space is constructed to convert the first position information and the second position information into a pose transformation matrix in a preset space;

[0038] Trajectory planning is performed based on the kinematic model and pose transformation matrix to generate an initial movement path; wherein, a velocity smoothing constraint of cubic spline interpolation is introduced into the configuration space corresponding to the initial movement path.

[0039] The collision detection information of the initial movement path is calculated based on the preset dynamic simulation, and the initial movement path is adjusted according to the collision detection information to generate the movement path; wherein, the collision detection information is calculated based on the collision probability, obstacle position and path energy consumption corresponding to the initial movement path;

[0040] The step of controlling the mobile robotic arm to apply the target inkjet pattern to the target area according to the movement path includes:

[0041] Obtain pressure feedback data from the end effector corresponding to the mobile robotic arm, and adjust the motion parameters of the mobile robotic arm based on the pressure feedback data;

[0042] The inkjet process of the mobile robotic arm is monitored in real time by a preset real-time image acquisition device. When a positional deviation and / or surface undulation change of the object to be inkjet is detected, the movement path is dynamically adjusted so as to control the mobile robotic arm to complete the inkjet process according to the dynamically adjusted motion parameters and movement path.

[0043] Preferably, the present invention further includes an intelligent inkjet control system for implementing the intelligent inkjet control method described above, the system comprising:

[0044] The feature construction module is used to acquire the surface contour data and color information of the object to be inkjetized based on the preset laser rangefinder, and to construct the object surface feature vector containing surface morphology features and material properties.

[0045] The region selection module is used to evaluate the region inkjetability of the surface feature vector of the object and obtain a suitable inkjet region distribution map including the probability distribution of inkjetable regions, so as to determine the target inkjet region based on the highest confidence value corresponding to the suitable inkjet region distribution map.

[0046] The parameter generation module is used to obtain the semantic information corresponding to the preset inkjet content, calculate the semantic element weights corresponding to the semantic information, and generate inkjet parameter constraints corresponding to the inkjet content based on the area range, semantic element weights, and semantic information corresponding to the target inkjet area; the inkjet parameter constraints include at least the droplet size threshold, information hierarchy presentation rules, inkjet pattern size parameters, and material adhesion firmness parameters.

[0047] The scheme formulation module is used to generate an intelligent inkjet scheme corresponding to the inkjet content based on the inkjet parameter constraints, so as to control the preset inkjet device to perform inkjet according to the intelligent inkjet scheme to form the target inkjet pattern.

[0048] The inkjet execution module is used to generate a preset movement path for the mobile robotic arm based on the first position information of the object to be inkjetted, the second position information of the preset inkjet device, and the area range corresponding to the target area, according to the cubic spline interpolation algorithm, so as to control the mobile robotic arm to apply the target inkjet pattern to the target area according to the movement path.

[0049] Preferably, the present invention further includes an intelligent inkjet control robot for performing the intelligent inkjet control method described above, the intelligent inkjet control robot comprising:

[0050] A laser rangefinder is used to acquire surface contour data and color information of the object to be inkjet printed.

[0051] Inkjet unit, used to perform inkjet operation;

[0052] A mobile robotic arm is used to apply the target inkjet pattern onto the object to be inkjet-ed;

[0053] A control device is configured to construct an object surface feature vector containing surface morphology features and material properties based on the surface contour data and color information; perform regional inkjet suitability assessment on the object surface feature vector to obtain an inkjet suitability region distribution map including the probability distribution of inkjetable regions, and determine the target inkjet region based on the highest confidence value corresponding to the inkjet suitability region distribution map; acquire semantic information corresponding to preset inkjet content, calculate the semantic element weights corresponding to the semantic information, and generate inkjet parameter constraints corresponding to the inkjet content based on the region range corresponding to the target inkjet region, the semantic element weights, and the semantic information; the inkjet parameter constraints... It includes at least droplet size thresholds, information hierarchy presentation rules, inkjet pattern size parameters, and material adhesion strength parameters; it generates an intelligent inkjet scheme corresponding to the inkjet content based on the inkjet parameter constraints, and controls a preset inkjet device to perform inkjet printing according to the intelligent inkjet scheme to form a target inkjet pattern; based on a cubic spline interpolation algorithm, it generates a preset movement path corresponding to a mobile robotic arm according to the first position information of the object to be inkjet printed, the second position information of the preset inkjet device, and the area range corresponding to the target area, and controls the mobile robotic arm to apply the target inkjet pattern to the target area according to the movement path.

[0054] Preferably, the present invention further includes a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the intelligent inkjet control method as described above.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] By acquiring surface contour data and color information of the object to be inkjetized using a laser rangefinder, a surface feature vector containing surface morphology and material properties is constructed. This vector accurately captures subtle differences on the object's surface, providing detailed foundational data for subsequent area inkjetability assessment. Based on this vector, the area inkjetability assessment determines the target inkjet area through the probability distribution of inkjetable regions, avoiding operation in unsuitable areas and reducing ink waste and unnecessary damage to the object's surface.

[0057] When processing inkjet content, this method extracts semantic information and calculates semantic element weights, then combines this with the range of the target inkjet area to generate inkjet parameter constraints. This ensures that the inkjet parameters match the semantic importance of the content. For example, for core information with high semantic weight, more reasonable droplet sizes and adhesion parameters can be set to ensure clear presentation on the object surface and prevent it from easily detaching. For secondary information, the pattern size can be adjusted according to the area range, ensuring information integrity while avoiding excessive space occupation. This parameter constraint generation method allows inkjet content to be presented in an appropriate form on different areas and materials, improving the effectiveness of information transmission.

[0058] The intelligent inkjet solution is generated based on the aforementioned parameter constraints, making the operation of the inkjet device more targeted. The moving robotic arm path generated by the cubic spline interpolation algorithm fully considers the positional relationship between the object to be inkjetized and the inkjet device, as well as the target area range. The path is smooth and the positioning is precise, reducing errors during the robotic arm's movement. This process achieves coordinated operation between the inkjet device and the moving robotic arm, ensuring that the target inkjet pattern can be accurately applied to the target area, even on complex surfaces such as curved or uneven surfaces, guaranteeing the integrity and consistency of the pattern.

[0059] The constraints involved in this method, such as droplet size thresholds and material adhesion parameters, can adapt to the surface characteristics of different materials. For smooth materials, parameters can be adjusted to enhance ink adhesion; for porous materials, droplet size can be controlled to prevent excessive penetration, thus forming stable inkjet patterns on various materials. Simultaneously, the application of information hierarchy presentation rules allows multi-layered semantic information to be displayed orderly within a limited area, avoiding information chaos and improving the overall aesthetics and readability of the pattern. Attached Figure Description

[0060] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent inkjet control method described in this invention.

[0061] Figure 2 A flowchart for constructing feature vectors for an object's surface;

[0062] Figure 3 A flowchart for regional sprayability assessment;

[0063] Figure 4 A flowchart defining the target inkjet area;

[0064] Figure 5 A flowchart generated for inkjet parameter constraints. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Please see Figures 1-5 This invention provides an intelligent inkjet control system, method, robot, and storage medium, wherein the method includes:

[0067] Step 1: Acquire the surface contour data and color information of the object to be inkjet-printed using a pre-set laser rangefinder. This data is used to construct a surface feature vector that includes surface morphology and material properties. The laser rangefinder can be a high-precision laser scanner that emits a laser beam to scan the object from all angles, acquiring the three-dimensional coordinate data of each point on the object's surface as surface contour data. Simultaneously, it collects the RGB color information of the object's surface using a built-in color sensor. The acquired surface contour data and color information are preprocessed to remove noise, followed by feature extraction and fusion to form a feature vector that comprehensively reflects the object's surface attributes.

[0068] Step 2: Assess the inkjetability of the object's surface feature vectors to obtain an inkjetable region distribution map, including the probability distribution of inkjetable areas. Determine the target inkjet region based on the highest confidence value corresponding to this distribution map. A specialized evaluation model is constructed to analyze different regions of the object's surface, determining whether each region is suitable for inkjet operation, and presenting the results as a probability distribution. The continuous regions with the highest confidence values ​​are selected from the inkjetable region distribution map and identified as the target inkjet regions.

[0069] Step 3: Obtain the semantic information corresponding to the preset inkjet content, calculate the semantic element weights corresponding to the semantic information, and generate inkjet parameter constraints corresponding to the inkjet content based on the area range corresponding to the target inkjet area, the semantic element weights, and the semantic information. The inkjet parameter constraints include at least droplet size thresholds, information hierarchy presentation rules, inkjet pattern size parameters, and material adhesion firmness parameters. The inkjet content may include text, patterns, etc., which are semantically parsed to extract key semantic information, such as the importance of text and the structural hierarchy of patterns, and the weights of each semantic element are calculated. Combined with the size, shape, and other range information of the target inkjet area, parameter constraints for regulating the inkjet process are generated.

[0070] Step 4: Generate an intelligent inkjet scheme corresponding to the inkjet content based on the inkjet parameter constraints. Control the preset inkjet device to perform inkjet printing according to the intelligent inkjet scheme to form the target inkjet pattern. Based on the inkjet parameter constraints, determine the working mode and inkjet sequence of the inkjet device, and formulate a detailed inkjet scheme. Control the printhead opening and closing, ink droplet ejection volume, etc., of the inkjet device according to this scheme to form the desired pattern in the target inkjet area.

[0071] Step 5: Based on the cubic spline interpolation algorithm, a preset movement path for the robotic arm is generated according to the first position information of the object to be inkjetized, the second position information of the preset inkjet device, and the area range corresponding to the target region. The robotic arm is then controlled to apply the target inkjet pattern to the target region according to the movement path. The first position information of the object to be inkjetized and the second position information of the inkjet device are obtained through a positioning device. Combined with the range of the target region, the movement path of the robotic arm is planned using the cubic spline interpolation algorithm. The robotic arm moves the inkjet device according to this path, ensuring that the target inkjet pattern is accurately applied to the target region.

[0072] Example 1:

[0073] In step 1, the surface contour data and color information of the object to be inkjetized are acquired using a preset laser rangefinder. This data is then used to construct a surface feature vector that includes surface morphology features and material properties. The specific process is as follows:

[0074] The laser ranging device can utilize a LiDAR with multi-line scanning capability. It emits multiple laser beams towards the surface of the object to be inkjet-printed, and calculates the three-dimensional coordinates of each sampling point on the object's surface using the laser reflection time difference, forming surface contour data in the form of a point cloud. Simultaneously, an integrated hyperspectral camera synchronously acquires color information from the object's surface, obtaining multi-channel color data including the visible spectrum. During the acquisition process, by setting the scanning frequency of the laser ranging device to match the camera's sampling frame rate, the precise spatial correspondence between the surface contour data and color information is ensured.

[0075] The color information is input into a pre-defined recurrent neural network. This recurrent neural network employs a bidirectional LSTM structure, consisting of an input layer, two bidirectional LSTM layers, and an output layer. The input layer converts the RGB channel data of the color information into a three-dimensional tensor, where the first dimension corresponds to the image height, the second dimension to the image width, and the third dimension to the three RGB color channels. The first bidirectional LSTM layer extracts features along the row direction of the tensor, capturing the color variation patterns in the horizontal direction; the second bidirectional LSTM layer extracts features along the column direction of the tensor, capturing the color distribution features in the vertical direction. After these two layers of processing, the output layer uses convolution operations to map the extracted features into color feature maps for multiple color channels, each feature map corresponding to a spatial distribution pattern of a specific color frequency band.

[0076] The surface contour data is input into a pre-defined point cloud processing network. This network includes a point cloud preprocessing module, a feature extraction module, and a morphological modeling module. The point cloud preprocessing module first denoises the raw point cloud data by removing discrete points too far from the mean using a statistical filtering algorithm, and then reduces the number of points while maintaining the overall structure of the point cloud using a voxel grid downsampling method. The feature extraction module uses a PointNet architecture, employing a multilayer perceptron to calculate features in the local neighborhood of each point, obtaining point feature vectors containing geometric features such as normals and curvature. The morphological modeling module inputs the point feature vectors into a graph convolutional network. Through message passing between graph nodes, it constructs the topological structure of the object's surface, calculates the spectral reflectance of each region, and derives material properties by combining geometric features, ultimately outputting surface morphological features containing both spectral reflectance and material properties.

[0077] The color feature map and the surface morphology feature are fused using feature vectors. The fusion process consists of two stages: spatial alignment and feature stitching. In the spatial alignment stage, the mapping relationship between the pixel coordinates of the color feature map and the point cloud coordinates of the surface morphology feature is calculated. Nearest neighbor interpolation is then used to project the surface morphology feature onto the two-dimensional plane of the color feature map, ensuring consistency in spatial dimension. In the feature stitching stage, the aligned color feature map and the surface morphology feature are stitched together along the channel dimension to form a high-dimensional fused feature vector.

[0078] The dimensionality of the fused feature vectors is compressed through normalization. Principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional fused feature vectors. The covariance matrix of the feature vectors is calculated, and their eigenvalues ​​and eigenvectors are solved. The top k eigenvectors with the largest eigenvalues ​​are selected to form a projection matrix, projecting the high-dimensional feature vectors into a low-dimensional space. During the projection process, local variance information of the feature vectors is preserved, ensuring that the dimensionality-reduced feature vectors still reflect the local morphological details of the object's surface. This ultimately generates object surface feature vectors with local morphological preservation. The dimensionality of these feature vectors is determined based on the accuracy requirements of the actual application scenario, typically controlled between 256 and 1024 dimensions to balance feature expressiveness and computational efficiency.

[0079] Throughout the processing, the parameters of the recurrent neural network and the point cloud processing network are pre-trained using a large-scale labeled dataset containing surface data of objects with different materials and shapes, along with their corresponding feature vector labels. The pre-training process employs a stochastic gradient descent algorithm with cross-entropy loss as the optimization objective. Through multiple iterations, the network parameters are adjusted to ensure the network can stably extract color and shape features that conform to actual physical properties. The parameters for feature fusion and normalization are dynamically adjusted based on the distribution characteristics of the input data, ensuring that the generated object surface feature vectors have a consistent scale and distribution range across different scenes.

[0080] Example 2:

[0081] In step 2, the surface feature vector of the object is evaluated for regional inkjetability to obtain an inkjetable region distribution map including the probability distribution of inkjetable regions. The specific process is as follows:

[0082] A deep region segmentation network is constructed. This network adopts the ResNet architecture's encoding and decoding structure. The encoding part consists of multiple convolutional modules, each containing a convolutional layer, a batch normalization layer, and an activation function. The convolutional layers use 3x3 kernels and perform convolution operations on the input object surface feature vectors through a sliding window to extract local features. The batch normalization layer standardizes the convolutional feature data to stabilize the data distribution. The ReLU activation function enhances the network's non-linear expressive power. A spatial attention module is introduced in the encoding stage, consisting of a channel attention submodule and a spatial attention submodule. The channel attention submodule converts the feature map into channel description vectors through global average pooling, and then calculates the importance weights of each channel through a fully connected layer. The spatial attention submodule performs max pooling and average pooling on the feature map along the channel dimensions, concatenates the results, and generates a spatial attention weight map through a convolutional layer. The channel weights are multiplied by the spatial weights to obtain the final attention weights, which are then weighted with the input feature vectors to highlight the feature information of key regions.

[0083] The decoding stage consists of multiple deconvolutional modules, each of which gradually restores the spatial resolution of the feature map through deconvolution operations. Dilated convolutional layers are embedded in the decoding stage. These layers have interleaved regions within the convolutional kernel, and the receptive field size is adjusted by changing the dilation rate. For regions with complex surface contours, a larger dilation rate is used to expand the receptive field and capture more contextual information; for relatively flat surfaces, a smaller dilation rate is used to improve the accuracy of local detail recognition. The alternation of dilated convolutional layers with regular convolutional layers allows the network to adapt to the contour features of different shaped regions, improving the accuracy of region segmentation.

[0084] The surface feature vector of the object is input into a deep region partitioning network. The feature vector is first processed by the convolutional and spatial attention modules in the encoding stage to progressively extract high-level features and compress the spatial dimension. Then, it enters the decoding stage, where deconvolution and dilated convolutional layers progressively restore the spatial size of the feature map. Simultaneously, feature information from different levels in the encoding stage is fused to enhance the richness of feature representation. The network's output layer uses the sigmoid activation function to convert the feature map values ​​into probability values ​​between 0 and 1, generating a printability probability map containing the printability of each pixel. The larger the value of each pixel in the probability map, the higher the probability that the location is suitable for inkjet printing.

[0085] The median filtering process is applied to the probability map. The window size for median filtering is determined based on the resolution of the probability map, typically a 3x3 or 5x5 window. For each pixel in the probability map, the probability values ​​of all pixels within its window are taken, these values ​​are sorted by size, and the median value is selected as the new value for that pixel. This method effectively removes isolated noise points in the probability map, avoiding the influence of single outliers on region identification. For pixels in edge regions, a mirror fill method is used to process the boundaries, ensuring the integrity of the edge contours after filtering.

[0086] After median filtering, the probability distribution in the inkjet readability probability map is smoother, and the probability values ​​of adjacent pixels change more continuously, forming an inkjet readability region distribution map with regional connectivity. In this distribution map, regions with high probability values ​​form continuous block structures, each block structure representing a potential inkjet-ready area. The boundaries between regions are distinguished by abrupt changes in probability values, resulting in clear boundary lines that facilitate subsequent identification and selection of target inkjet regions.

[0087] The training process of the deep region segmentation network utilizes a dataset of inkjet-friendly regions labeled with various object surfaces. This dataset covers a variety of materials, including metal, plastic, and wood, as well as surface morphology such as flat, curved, and uneven surfaces. During training, the cross-entropy loss between the predicted inkjet-friendly probability map and the labeled inkjet-friendly region map is used as the optimization objective, and the Adam optimization algorithm is employed to adjust the network parameters. Through multiple rounds of iterative training, the network accurately learns the mapping relationship between different surface features and inkjet-friendly properties, ensuring that the generated inkjet-friendly region distribution map conforms to actual inkjet requirements. The median filter parameters are adjusted based on the performance on the validation set during training to achieve optimal noise removal and region preservation.

[0088] Example 3:

[0089] In step 2, the target inkjet region is determined based on the highest confidence value corresponding to the suitable inkjet region distribution map. The specific process is as follows:

[0090] An evaluation index system is constructed based on the maximum coverage area, minimum surface roughness variance, and maximum edge sharpness index corresponding to the target inkjet region. The maximum coverage area refers to the area occupied by the candidate region on the object's surface, obtained by calculating the two-dimensional projected area enclosed by the region boundaries. The curvature of the object's surface must be considered during area calculation; the curved surface region is unfolded before surface accumulation. The minimum surface roughness variance is obtained by averaging the sum of squared deviations of the roughness values ​​at each point within the region from the region's average roughness value. The roughness value is derived from the morphological feature parameters in the object's surface feature vector and reflects the smoothness of the region's surface. The maximum edge sharpness index is determined by analyzing the rate of change of grayscale values ​​of pixels at the region's boundary. A larger rate of change indicates a sharper edge. Noise interference at the boundary must be eliminated during calculation, and edge pixels are smoothed before gradient calculation.

[0091] The suitable inkjet region distribution map is optimized based on a pre-defined evaluation index system and a particle swarm optimization algorithm to obtain multiple candidate inkjet regions. During the initialization of the particle swarm optimization algorithm, the position information of each candidate region is used as the initial position of the particles, and the particle dimension is determined by the boundary coordinate parameters of the region. During algorithm iteration, each particle adjusts its flight speed and direction based on its own historical best position and the global best position of the swarm. An inertia weight factor is introduced when updating the speed, and this weight is gradually reduced as the number of iterations increases, allowing the particles to maintain their exploration ability in the early stages and enhance their local search ability in the later stages. The particle fitness function is composed of a weighted average of three indicators from the evaluation index system. The weights are set according to the relative importance of each indicator in different application scenarios, and the quality of candidate regions is judged by the fitness value. After a pre-defined number of iterations, the regions corresponding to multiple particles with high fitness values ​​are selected as candidate inkjet regions.

[0092] The MOEA / D algorithm is used to perform non-dominated ranking of multiple candidate inkjet regions. During this ranking process, each candidate region is considered a solution. If solution A is superior to solution B in all evaluation metrics, then solution A dominates solution B, and solution B is marked as a dominated solution. Through multiple rounds of comparison, all candidate regions are divided into different non-dominated levels. The smaller the level number, the better the overall performance of the candidate regions at that level. After the non-dominated ranking is completed, the crowding degree of the candidate regions within each level is calculated. Crowding degree reflects the density of candidate regions in the target space; regions with higher crowding degree are retained to maintain the diversity of candidate regions.

[0093] The comprehensive score for each candidate inkjet region after ranking is calculated using the Analytic Hierarchy Process (AHP). AHP first constructs a judgment matrix, where matrix elements represent the relative importance of two evaluation indicators, with importance levels ranging from 1 to 9, where higher levels indicate greater relative importance. A consistency check is performed on the judgment matrix by calculating consistency indices and consistency ratios to ensure its rationality. If the consistency ratio exceeds a preset threshold, the values ​​of the judgment matrix elements are readjusted. After the consistency check passes, the eigenvector corresponding to the largest eigenvalue of the judgment matrix is ​​calculated and used as the weight coefficient for each evaluation indicator. The formula for calculating the comprehensive score is:

[0094] S=α×S1+β×S2+γ×S3

[0095] Where S represents the overall score of the candidate inkjet region, α, β, and γ represent the weighting coefficients of the maximum coverage area, minimum surface roughness variance, and maximum edge sharpness index, respectively, S1 represents the standardized value of the maximum coverage area, S2 represents the standardized value of the minimum surface roughness variance, and S3 represents the standardized value of the maximum edge sharpness index. During standardization, the original values ​​of each index are mapped to the range of 0 to 1. For the maximum coverage area and the maximum edge sharpness index, forward standardization is used, meaning the larger the index value, the closer the standardized value is to 1; for the minimum surface roughness variance, backward standardization is used, meaning the smaller the index value, the closer the standardized value is to 1.

[0096] Among multiple candidate inkjet regions, continuous regions with a comprehensive score greater than a threshold are selected as target inkjet regions. The threshold is determined based on actual inkjet requirements, by analyzing the comprehensive score distribution of qualified regions in historical inkjet data. Typically, the average comprehensive score of all candidate regions is used as the initial threshold, which is then adjusted according to the continuity requirements of the regions. Continuous regions are determined by analyzing the connectivity of pixels within the region, using the 8-neighborhood connectivity criterion: if two adjacent pixels are both suitable inkjet regions and their comprehensive scores are both greater than the threshold, then these two pixels are considered to belong to the same continuous region. Boundary extraction is performed on qualified continuous regions, and the contour lines are obtained using edge detection algorithms. The contour lines are simplified using polygon fitting to reduce the computational load of subsequent path planning. If multiple qualified continuous regions exist, the region with the highest comprehensive score is selected as the target inkjet region. If the comprehensive scores of all candidate regions are less than the threshold, the region suitability assessment is repeated, the assessment parameters are adjusted, and a new suitable inkjet region distribution map is generated until a suitable target inkjet region is determined.

[0097] Throughout the process, the number of iterations in the Particle Swarm Optimization (PSO) algorithm, the population size in the MOEA / D algorithm, and the dimension of the judgment matrix in the Analytic Hierarchy Process (AHP) are dynamically adjusted based on the number and complexity of the candidate regions. For objects with simple surface morphology, the number of iterations and the population size can be appropriately reduced to improve processing efficiency; for objects with complex surface structures, the number of iterations and the population size are increased to ensure that the selected target inkjet regions meet the actual inkjet requirements. In the calculation of the overall score, the weighting coefficients can be dynamically adjusted according to the characteristics of the inkjet content. For example, for inkjet content that requires large-area presentation, the weighting coefficient of the maximum coverage area is increased; for inkjet content that requires sharp edges, the weighting coefficient of the edge sharpness index is increased.

[0098] Example 4:

[0099] In step 3, the semantic element weights corresponding to the semantic information are calculated to generate inkjet parameter constraints corresponding to the inkjet content based on the region range corresponding to the target inkjet region, the semantic element weights, and the semantic information. The specific process is as follows:

[0100] The semantic association graph corresponding to the semantic information is extracted based on a pre-defined LSTM model. If the inkjet content is a combination of text and image information, the text part is first segmented into words, and then the grammatical attributes of each word are determined through part-of-speech tagging. The image part is contour extracted to identify the main constituent elements and the connection relationships between them. The input to the LSTM model is the pre-processed text segmentation sequence and the feature vector of the image elements. The model captures the dependency relationships between elements in the sequence through multiple layers of hidden connections, such as the relationship between modifiers and head words in the text, and the spatial position relationship between main elements and auxiliary elements in the image. The semantic association graph output by the model represents semantic elements with nodes. The node attributes include element type (text / image), feature value (such as part of speech of text, shape of image elements), and directed edges represent the relationship type between elements (such as modification relationship, inclusion relationship), with the edge weight reflecting the relationship strength.

[0101] The importance score of each semantic node in the semantic association graph is calculated based on a pre-defined graph neural network and used as the weight of the semantic element. The input to the graph neural network is the adjacency matrix and the node feature matrix of the semantic association graph. The adjacency matrix records the connection relationships between nodes, and the node feature matrix contains the attribute information of each node. The network updates the node features through graph convolutional layers. The new feature of each node is obtained by weighted summation of its own features and the features of its neighboring nodes, and the weights are dynamically adjusted according to the association strength of the edges. After multiple layers of graph convolutional operations, the features of all nodes are aggregated into graph-level features through a global pooling layer, and then the importance score of each node is output through a fully connected layer. In the scoring process, for text elements, the focus is on their coreness in semantic expression; for pattern elements, the focus is on their prominence in visual presentation. The final importance score is the weight of each semantic element.

[0102] The region range parameters are transformed into printhead arrangement constraints, and semantic element weights are mapped into weighted coefficients for information hierarchy presentation rules. Geometric parameters within the region range include the length, width, area, and proportional relationships of each dimension of the target inkjet region. These parameters determine the number and arrangement of printheads; for example, when the region width is large, multiple printheads are arranged side-by-side to cover the entire width. For morphological parameters, if the displayed region is irregularly shaped, the printhead arrangement constraints must include the printhead turning angle range to ensure the printheads are always aligned with the region's center. Edge smoothness, among edge characteristic parameters, determines the inkjet accuracy requirements when the printhead approaches the region's edge; smoother edges require higher accuracy. Surface hardness, among material characteristic parameters, affects the distance constraint between the printhead and the object's surface; materials with lower hardness require increased printhead distance to avoid contact damage. Spatial position and orientation parameters determine the initial positioning coordinates and working plane of the printhead. The connection relationship with other regions, among topology parameters, determines the transition method when the printhead switches between regions. For stability parameters, if the displayed region has a risk of deformation, the printhead arrangement constraints must include a pressure adjustment range. The higher the semantic element weight, the larger its corresponding weighting coefficient. In the information hierarchy presentation rules, elements with larger weighting coefficients will use higher ink density or more eye-catching colors.

[0103] The dynamic programming method is used to obtain the inkjet parameter constraints corresponding to the inkjet content based on printhead arrangement constraints and weighting coefficients. The dynamic programming method decomposes the inkjet process into multiple continuous inkjet segments, each corresponding to a sub-region within the target area. For each sub-region, the printhead combination and inkjet range are determined according to the printhead arrangement constraints, and the inkjet resources for each semantic element in that segment, such as the number of droplets and the ejection time, are allocated using weighting coefficients. In determining the droplet size threshold, the material characteristics of the sub-region and the weights of the semantic elements are considered comprehensively. Larger droplet sizes are used for regions with high-weight elements and good material adhesion; smaller droplet sizes are used for regions with low-weight elements and poor material adhesion. The information hierarchy presentation rules are sorted according to the weighting coefficients to determine the inkjet order of each element, with higher-weight elements being inkjet-first to ensure their integrity. The inkjet pattern size parameter is determined based on the proportional relationship between the region's geometric dimensions and the semantic elements, ensuring the pattern maintains an appropriate scaling ratio within the region. The material adhesion strength parameter refers to the surface roughness in the material characteristic parameters; regions with higher roughness require an increased proportion of adhesive for the droplets to enhance adhesion. Through multi-stage decision-making via dynamic programming, inkjet parameter constraints containing all the above parameters are ultimately generated, ensuring that the inkjet process is adapted to the characteristics and semantic information requirements of the target area.

[0104] Example 5:

[0105] In step 4, an intelligent inkjet solution corresponding to the inkjet content is generated based on the inkjet parameter constraints. The specific process is as follows:

[0106] A smart inkjet architecture based on a reinforcement learning policy network is constructed. The policy network of the reinforcement learning policy network employs a multi-head attention mechanism, a multi-scale feature fusion network. This mechanism includes multiple parallel attention heads, each focusing on different feature dimensions within the inkjet content and inkjet parameter constraints. For example, some attention heads focus on the matching relationship between droplet size thresholds and material properties, while others focus on the adaptability of information hierarchy presentation rules and regional ranges. The multi-scale feature fusion network extracts feature information at different scales using convolutional kernels of varying sizes. Small-sized kernels capture local detail features, while large-sized kernels extract global structural features. These features are fused at multiple levels to form a comprehensive feature set that fully reflects the inkjet scene. The value network of the reinforcement learning policy network incorporates a simulation rendering module. This module simulates the diffusion and adhesion process of ink droplets after ejection using a physics engine, generating virtual inkjet effect images based on inkjet parameter constraints. These images include droplet distribution density, color overlay effects, and adhesion to object surfaces. The policy network parameters of the reinforcement learning policy network are optimized based on policy gradient. During the optimization process, the corresponding reward function for reinforcement learning is generated based on information entropy, visual consistency, and material matching degree. Information entropy is obtained by calculating the uncertainty of the ink droplet size distribution in the inkjet scheme; the more uniform the distribution, the greater the information entropy. Visual consistency is obtained by analyzing the color harmony and pattern integrity in the virtual inkjet effect image. Material matching degree is determined by comparing the degree of fit between the ink droplet adhesion parameters and the material characteristic parameters.

[0107] A style parameter vector is generated based on preset visual specifications and inkjet parameter constraints. The visual specifications cover color matching range, font edge processing methods, and pattern line thickness standards. These contents are mapped to ink droplet size thresholds and inkjet pattern size parameters in the inkjet parameter constraints, and converted into numerical style parameters. For example, the color matching range is converted into a numerical range in the RGB color space, and the font edge processing method is converted into a numerical index of edge blur. After forming the style parameter vector, it is input into the feature space corresponding to the intelligent inkjet architecture, so that the architecture can understand and follow the preset visual requirements.

[0108] The inkjet content is input into the intelligent inkjet architecture. The architecture uses a policy network to comprehensively analyze the inkjet content, style parameter vectors, and inkjet parameter constraints to generate a preliminary inkjet solution. This solution includes parameters such as printhead activation timing, droplet ejection frequency, and travel speed. The value network evaluates the preliminary solution through a simulation rendering module, outputting a reward function value. The policy network adjusts its parameters based on this value, iterating repeatedly until a stable intelligent inkjet solution with a stable reward function value is generated.

[0109] In step 5, based on the cubic spline interpolation algorithm, a preset movement path for the mobile robotic arm is generated according to the first position information of the object to be inkjetized, the preset second position information of the inkjet device, and the area range corresponding to the target area. The specific process is as follows:

[0110] A kinematic model based on Cartesian space is constructed. This model, based on the Cartesian coordinate system, includes parameters such as the length, connection relationship, and range of motion of each joint of the moving robotic arm. The first position information of the object to be inkjetized is converted into the origin and attitude matrix of the object in Cartesian space. The second position information of the inkjet device is converted into the initial coordinates and attitude matrix of the end effector. The kinematic model is used to calculate the pose transformation matrix between the two, which contains translational and rotational components, representing the relationship between position and attitude changes, respectively.

[0111] Trajectory planning is performed based on the kinematic model and pose transformation matrix to generate an initial movement path. The initial path consists of multiple path points, with adjacent path points connected by cubic spline curves. The first and second derivatives of these curves are continuous at the path points, ensuring the smoothness of the robotic arm's movement. A velocity smoothing constraint based on cubic spline interpolation is introduced into the configuration space corresponding to the initial movement path. This constraint includes maximum velocity thresholds, acceleration thresholds, and jerk thresholds. By limiting the rate of velocity change at the path points, sudden stops and starts of the robotic arm during movement are prevented.

[0112] The initial movement path collision detection information is calculated based on a preset dynamic simulation. The dynamic simulation simulates the force conditions of each joint of the mobile robotic arm as it moves along the initial path, as well as the relative positional relationship between the movement trajectory and the surrounding environment. The collision probability is obtained by calculating the spatial overlap between the robotic arm model and the obstacle model; the higher the overlap, the greater the collision probability. The obstacle position is determined by real-time environmental point cloud data acquired by a laser rangefinder. The path energy consumption is calculated by accumulating the product of the driving torque and the movement angle of each joint during the movement. The initial movement path is adjusted based on the collision detection information. If the collision probability exceeds the set value, obstacles are avoided by adding path points or changing the path direction. If the path energy consumption is too high, the distribution of path points is optimized, and the movement distance is shortened to reduce energy consumption, ultimately generating an adjusted movement path.

[0113] The mobile robotic arm, guided by the movement path, applies the target inkjet pattern to the target area. The specific process is as follows:

[0114] The system acquires pressure feedback data from the end effector of the mobile robotic arm. The pressure sensor installed on the end effector collects the pressure value when it comes into contact with the object surface in real time. The motion parameters of the mobile robotic arm are adjusted according to the changes in the pressure value. When the pressure value exceeds the set range, the feed speed of the end effector is reduced; when the pressure value is lower than the set range, the feed speed is increased so that the pressure is always kept in a suitable range.

[0115] The inkjet process of the mobile robotic arm is monitored in real time by a pre-set real-time image acquisition device. This device uses a high-speed industrial camera to capture images of the inkjet area at a fixed frame rate. Image recognition algorithms analyze the deviation between the object's position in the image and the pre-set first position information, as well as the surface undulations. When a positional deviation is detected, the pose transformation matrix is ​​recalculated to dynamically correct the movement path. When surface undulations are detected, the height of the end effector is adjusted according to the amplitude of the undulations to ensure a stable inkjet distance. Based on the dynamically adjusted motion parameters and movement path, the mobile robotic arm is controlled to complete the inkjet process.

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0117] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent inkjet control method, characterized by, The application relates to a method for generating an intelligent inkjet scheme for an object, and belongs to the technical field of intelligent manufacturing. The method comprises the following steps: acquiring surface profile data and color information of an object to be inkjetted by using a preset laser ranging device, and using the surface profile data and the color information to construct an object surface feature vector containing surface morphology features and material characteristics; performing region adaptability evaluation on the object surface feature vector to acquire an inkjet adaptability region distribution map including a probability distribution of an inkjettable region, and determining a target inkjet region according to a highest confidence value corresponding to the inkjet adaptability region distribution map; acquiring semantic information corresponding to preset inkjet content, calculating a semantic element weight corresponding to the semantic information, and generating inkjet parameter constraints corresponding to the inkjet content according to a region range corresponding to the target inkjet region, the semantic element weight and the semantic information; the inkjet parameter constraints at least include an ink droplet size threshold, an information level presentation rule, an inkjet pattern size parameter and a material adhesion firmness parameter; generating an intelligent inkjet scheme corresponding to the inkjet content according to the inkjet parameter constraints, and controlling a preset inkjet device to perform inkjetting according to the intelligent inkjet scheme to form a target inkjet pattern; generating a moving path corresponding to a preset moving mechanical arm according to first position information corresponding to the object to be inkjetted, second position information of the preset inkjet device and a region range corresponding to the target region based on a cubic spline interpolation algorithm, and controlling the moving mechanical arm to act the target inkjet pattern on the target region according to the moving path; The method comprises the following steps: acquiring surface profile data and color information of an object to be inkjetted by using a preset laser ranging device, and using the surface profile data and the color information to construct an object surface feature vector containing surface morphology features and material characteristics; inputting the color information into a preset recurrent neural network to extract color feature maps of multiple color channels; inputting the surface profile data into a preset point cloud processing network to model the surface morphology of the profile data, and outputting surface morphology features including spectral reflectance and material characteristics; performing feature vector fusion on the color feature maps and the surface morphology features, performing dimension compression on the fused feature vectors through normalization processing, and generating an object surface feature vector with local morphology retention; The method comprises the following steps: constructing a deep region division network; the deep region division network adopts a coding-decoding structure of a ResNet architecture, introduces a spatial attention module in an encoding stage to perform importance weighting on the object surface feature vector, and embeds a hollow convolutional layer in a decoding stage to dynamically adjust a receptive field to adapt to profile features of different morphology regions; inputting the object surface feature vector into the deep region division network to output an inkjet adaptability probability map containing inkjet adaptability degrees of each pixel point; performing median filter processing on the inkjet adaptability probability map to eliminate discrete noise points, and generating an inkjet adaptability region distribution map with region connectivity. The method comprises the following steps: constructing an evaluation index system according to a maximum coverage area, a minimum surface roughness variance and a maximum edge sharpness index corresponding to the target inkjet region; 2. The intelligent inkjet control method of claim 1, wherein, ​ ​ According to the preset evaluation index system and the particle swarm optimization algorithm, the suitable spraying area distribution map is optimized to obtain a plurality of candidate inkjet regions; According to the MOEA / D algorithm, the plurality of candidate inkjet regions are non-dominated sorted, and the comprehensive score corresponding to each candidate inkjet region after sorting is calculated according to the analytic hierarchy process; In the plurality of candidate inkjet regions, a continuous region with a comprehensive score greater than a threshold value is obtained as the target inkjet region.

3. The intelligent inkjet control method of claim 1, wherein, The semantic element weight corresponding to the semantic information is calculated to generate the inkjet parameter constraint corresponding to the inkjet content according to the region range corresponding to the target inkjet region, the semantic element weight and the semantic information, including: According to a preset LSTM model, a semantic association graph corresponding to the semantic information is extracted; Based on a preset graph neural network, the importance score of each semantic node of the semantic association graph is calculated as the semantic element weight; The region range parameter corresponding to the region range is converted into a nozzle arrangement constraint, and the semantic element weight is mapped into a weighted coefficient of information level presentation rule; the region range parameter includes one or more of geometric size parameter, morphological feature parameter, edge feature parameter, material characteristic parameter, spatial position and direction parameter, topological parameter and stability parameter; Based on a dynamic programming method, the inkjet parameter constraint corresponding to the inkjet content is obtained according to the nozzle arrangement constraint and the weighted coefficient.

4. The intelligent inkjet control method of claim 1, wherein, The intelligent inkjet scheme corresponding to the inkjet content is generated according to the inkjet parameter constraint, including: An intelligent inkjet architecture based on a reinforcement learning policy network is constructed; the policy network of the reinforcement learning policy network adopts a multi-scale feature fusion network with a multi-head attention mechanism, and the value network of the reinforcement learning policy network introduces a simulation rendering module to simulate the actual inkjet effect; the policy network parameters of the reinforcement learning policy network are optimized based on policy gradient, and the reward function corresponding to the reinforcement learning is generated according to information entropy, visual coordination degree and material matching degree; According to a preset visual specification and inkjet parameter constraint, a style parameter vector is generated, and the style parameter vector is input into a feature space corresponding to the intelligent inkjet architecture; The inkjet content is input into the intelligent inkjet architecture to generate the intelligent inkjet scheme.

5. The intelligent inkjet control method of claim 1, wherein, Based on a cubic spline interpolation algorithm, a moving path corresponding to a preset moving mechanical arm is generated according to first position information corresponding to a to-be-inkjet object, second position information of a preset inkjet device and a region range corresponding to a target region, including: A kinematics model based on Cartesian space is constructed to convert the first position information and the second position information into a pose transformation matrix in a preset space; According to the kinematics model and the pose transformation matrix, a trajectory planning is performed to generate an initial moving path; wherein a velocity smoothing constraint of cubic spline interpolation is introduced in a configuration space corresponding to the initial moving path; According to a preset dynamics simulation, collision detection information of the initial moving path is calculated to adjust the initial moving path according to the collision detection information to generate the moving path; wherein the collision detection information is calculated according to the collision probability, obstacle position and path energy consumption corresponding to the initial moving path. The method comprises the following steps: acquiring pressure feedback data of an end effector corresponding to the mobile manipulator, and adjusting motion parameters of the mobile manipulator according to the pressure feedback data; monitoring an inkjet process of the mobile manipulator in real time according to a preset real-time image acquisition device, and dynamically adjusting the movement path when detecting that a position deviation and / or surface undulation change of an object to be inkjetted occurs, so as to control the mobile manipulator to complete the inkjet process according to the dynamically adjusted motion parameters and movement path.

6. An intelligent inkjet control system for implementing the intelligent inkjet control method according to any one of claims 1 to 5, characterized in that The method comprises the following steps: a feature construction module is configured to acquire surface profile data and color information of an object to be inkjetted according to a preset laser ranging device, and to construct an object surface feature vector containing surface morphology features and material characteristics; a region selection module is configured to perform region suitability evaluation on the object surface feature vector, to acquire a suitable region distribution map including a probability distribution of inkjettable regions, and to determine a target inkjet region according to a highest confidence value corresponding to the suitable region distribution map; a parameter generation module is configured to acquire semantic information corresponding to preset inkjet content, to calculate semantic element weights corresponding to the semantic information, and to generate inkjet parameter constraints corresponding to the inkjet content according to a region range corresponding to the target inkjet region, the semantic element weights, and the semantic information; the inkjet parameter constraints at least include an ink droplet size threshold, an information level presentation rule, an inkjet pattern size parameter, and a material adhesion firmness parameter; a scheme formulation module is configured to generate an intelligent inkjet scheme corresponding to the inkjet content according to the inkjet parameter constraints, and to control a preset inkjet device to perform inkjetting according to the intelligent inkjet scheme, so as to form a target inkjet pattern; an inkjet execution module is configured to generate a movement path corresponding to a preset mobile manipulator according to first position information corresponding to an object to be inkjetted, second position information of a preset inkjet device, and a region range corresponding to a target region based on a cubic spline interpolation algorithm, and to control the mobile manipulator to act the target inkjet pattern on the target region according to the movement path.

7. An intelligent inkjet control robot for performing the intelligent inkjet control method according to any one of claims 1 to 5, characterized in that, The method comprises the following steps: a laser ranging device is configured to acquire surface profile data and color information of an object to be inkjetted; an inkjet device is configured to perform inkjetting operation; a mobile manipulator is configured to act a target inkjet pattern on an object to be inkjetted; a control device is configured to construct an object surface feature vector containing surface morphology features and material characteristics according to the surface profile data and color information, to perform region suitability evaluation on the object surface feature vector, to acquire a suitable region distribution map including a probability distribution of inkjettable regions, and to determine a target inkjet region according to a highest confidence value corresponding to the suitable region distribution map. The semantic information corresponding to the preset inkjet content is acquired, the semantic element weight corresponding to the semantic information is calculated, the inkjet parameter constraint corresponding to the inkjet content is generated according to the region range corresponding to the target inkjet region, the semantic element weight and the semantic information, the inkjet parameter constraint at least includes an ink drop size threshold, an information layer level presentation rule, an inkjet pattern size parameter and a material adhesion firmness parameter, the intelligent inkjet scheme corresponding to the inkjet content is generated according to the inkjet parameter constraint, and the preset inkjet device is controlled to perform inkjet according to the intelligent inkjet scheme to form a target inkjet pattern. According to the first position information corresponding to the object to be inkjetted, the second position information of the preset inkjet device and the region range corresponding to the target region, a moving path corresponding to the preset moving mechanical arm is generated based on a cubic spline interpolation algorithm, and the moving mechanical arm is controlled to act the target inkjet pattern on the target region according to the moving path.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a stored computer program, and when a device where the computer readable storage medium is located executes the computer program, the intelligent inkjet control method in any one of claims 1 to 5 is implemented.

Images