Palletizer control system

Abstract

The present application discloses a palletizer control system, which, during the palletizing process, first obtains a surface state image of a mold base component through an industrial camera, and uses an image recognition algorithm based on deep learning to extract local implicit features and global semantic features of the mold base component surface state to more comprehensively describe the state of the mold base component. Feature selection is then performed on the extracted mold base component surface state features to remove redundancy in the features, thereby improving the characterization capability of the mold base component surface state features. Further, based on the mold base component surface state features after feature selection, defects of the mold base component are identified. In this way, the system can improve the accuracy and reliability of mold base component quality detection while ensuring efficient production, thereby improving the intelligence level of the palletizer control system.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to a palletizer control system. Background Art

[0002] In the field of industrial automation, palletizers, an integral part of the logistics and manufacturing processes, undertake the crucial task of stacking and removing products according to specific patterns. As the manufacturing industry continues to demand higher efficiency and quality, traditional palletizing methods are increasingly showing their limitations, particularly when handling complex shapes and high-precision products such as pallet components. Traditional palletizer control systems often rely on pre-set action sequences to execute tasks, which not only limits the system's flexibility but also makes it prone to operational errors when dealing with components with non-standard dimensions or abnormal surface conditions, leading to reduced product quality or production process interruptions.

[0003] Patent CN112850172A provides a control system and method for a palletizer. The palletizing process is a highly integrated and intelligent system that exchanges information with the curing kiln and assembly line systems via a controller and communication module. During this process, the palletizer executes actions based on task information received from these two systems, ensuring the smooth flow of pallet components between the curing kiln and the assembly line. Specifically, when a pallet needs to be stored in the curing kiln, the palletizer first transfers the pallet from the assembly line to its pallet. Then, through precise control of the layers and rows, it moves the pallet to the designated destination bin, completing a series of operations: opening the bin door, pushing the pallet into the curing kiln, and closing the bin door. Conversely, when a cured pallet needs to be removed, these steps are reversed, ultimately returning the pallet to the assembly line. This system not only enables the automated flow of pallet components between the curing kiln and the assembly line, but also introduces intelligent control methods to optimize the storage and retrieval process. However, in this highly integrated and automated process, if defective components are not effectively identified and removed, they may enter subsequent processes, leading to more serious quality issues and even affecting the overall performance of the finished product. Therefore, ensuring the quality of each mold base component has become a key point that cannot be ignored.

[0004] Therefore, an optimized palletizer control system is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. An embodiment of the present application provides a palletizer control system, which, during the palletizing process, first obtains a surface state image of a mold base component of a mold base component through an industrial camera, and uses an image recognition algorithm based on deep learning to extract local implicit features and global semantic features of the surface state of the mold base component to more comprehensively describe the state of the mold base component, and then performs feature selection on the extracted surface state features of the mold base component to remove redundancy in the features, thereby improving the characterization capability of the surface state features of the mold base component, and further realizes defect recognition of the mold base component based on the surface state features of the mold base component after feature selection. In this way, the system can improve the accuracy and reliability of mold base component quality detection while ensuring efficient production, thereby improving the intelligence level of the palletizer control system.

[0006] According to one aspect of the present application, a palletizer control system is provided, comprising: a controller and a communication module; the controller communicates with a curing kiln system and an assembly line system via the communication module; the controller is further configured to obtain task information of the curing kiln system and the assembly line system via the communication module, and control the operation of the palletizer based on the task information to enable the transfer of pallet components between the curing kiln and the assembly line, characterized in that it further comprises:

[0007] an industrial camera for capturing an image of a surface state of the die plate component, wherein the die plate component is in a state of being transferred between the curing kiln and the assembly line;

[0008] The communication module is further used to transmit the surface state image of the die table component to the controller;

[0009] The controller is further configured to perform image recognition on the surface state image of the die table component to obtain a recognition result, wherein the recognition result is used to indicate whether the die table component has a defect;

[0010] The controller is further configured to generate a palletizing abnormality warning prompt in response to the identification result indicating that the mold platform component has a defect.

[0011] Compared with the existing technology, the present application provides a palletizer control system, which, during the palletizing process, first obtains the surface state image of the mold base component through an industrial camera, and uses an image recognition algorithm based on deep learning to extract the local implicit features and global semantic features of the mold base component surface state to more comprehensively describe the state of the mold base component. It then performs feature selection on the extracted mold base component surface state features to remove redundancy in the features, thereby improving the characterization capability of the mold base component surface state features, and further realizes the defect identification of the mold base component based on the mold base component surface state features after feature selection. In this way, the system can improve the accuracy and reliability of mold base component quality detection while ensuring efficient production, thereby improving the intelligence level of the palletizer control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 is a block diagram of a palletizer control system according to an embodiment of the present application;

[0014] Figure 2 Schematic diagram of data flow of a palletizer control system according to an embodiment of the present application;

[0015] Figure 3 4 is a block diagram of a controller in a palletizer control system according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0017] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0018] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0019] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0020] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0021] It should be understood that if defective components are not effectively identified and removed during the palletizing process, these defective products may enter subsequent processes, leading to more serious quality problems and even affecting the overall performance of the finished product. Therefore, ensuring the quality of each pallet component has become a key point that cannot be ignored.

[0022] Based on this, the technical solution of the present application first uses an industrial camera to capture images of the surface status of the mold pallet components as they flow between the curing kiln and the assembly line. These images are then transmitted to a controller in real time via a communication module. The controller then performs image recognition processing on the received images of the mold pallet components to analyze whether the components are defective and obtain a corresponding recognition result. In response to the controller determining that the mold pallet components are defective, a stacking anomaly warning is generated to notify the operator or system to take appropriate measures to ensure that the problematic mold pallet components do not continue to participate in subsequent processes, thereby ensuring product quality and the smooth operation of the production process.

[0023] Specifically, in the technical solution of the present application, a palletizer control system is proposed. Figure 1 4 is a block diagram of a palletizer control system according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the palletizer control system according to the embodiment of the present application. Figure 1 and Figure 2As shown, a palletizer control system 300 according to an embodiment of the present application includes: an industrial camera 310 for capturing a surface state image of the mold base component, where the mold base component is in a state of flowing between the curing kiln and the assembly line; a communication module 320 for transmitting the surface state image of the mold base component to the controller; and a controller 330, which communicates with the curing kiln system and the assembly line system through the communication module.

[0024] In particular, the industrial camera 310 is used to capture the surface state image of the mold base component of the mold base component, and the mold base component is in a state of flowing between the curing kiln and the assembly line. In one example, the industrial camera can be installed at a key position, such as the exit of the curing kiln or the entrance of the assembly line. The purpose of this setting is to ensure that whenever the mold base component passes through these fixed points, its surface state can be accurately photographed. In order to accurately control the timing of shooting, the industrial camera will be connected to the sensor network in the system, such as a photoelectric sensor or other types of proximity switches. Once the sensor detects that the mold base component approaches or reaches the specified position, it will trigger the industrial camera to start working, thereby obtaining high-quality image data.

[0025] Specifically, the communication module 320 is used to transmit images of the mold base component's surface condition to the controller. In one example, the industrial camera must first be connected to the communication module via an appropriate interface (such as GigE Vision, Camera Link, etc.). This hardware connection ensures stable and high-speed transmission of image data from the industrial camera to the communication module. Given the complexity and variability of industrial environments, selecting the appropriate interface is crucial to ensuring the quality of image data transmission. For example, in a large factory environment, an Ethernet interface may be the best choice, offering long-distance transmission capabilities, good anti-interference performance, and ease of integration with other network devices. Next, to ensure smooth transmission of image data to the controller, the communication module must support a data transmission protocol compatible with the industrial camera. If the industrial camera uses the TCP / IP protocol for network communication, the communication module must also be capable of processing this data stream. In practice, this means that the communication module must not only be able to receive data packets from the industrial camera but also accurately forward this data to the controller. Furthermore, after the industrial camera completes a capture, it generates an image file containing the mold base component's surface condition. These image files are encoded into a format suitable for network transmission and packaged by the communication module for transmission to the controller. This process may involve the application of compression algorithms to reduce bandwidth usage and speed up transmission. Especially with high-resolution images, effective compression techniques can significantly improve transmission efficiency without significantly compromising image quality. Furthermore, given the critical importance of image data, the communication module should also include error detection and correction mechanisms to ensure high data integrity even under poor network conditions.

[0026] Specifically, the controller 330 communicates with the curing kiln system and the assembly line system via the communication module; and is configured to obtain task information from the curing kiln system and the assembly line system via the communication module, and control the operation of the palletizer based on the task information to enable the transfer of mold pallet components between the curing kiln and the assembly line. Specifically, the controller 330 includes: receiving first status information from the curing kiln system via the communication module to form a fetch task queue; receiving second status information from the assembly line system via the communication module to form a store task queue; determining operating status information based on the first and second status information to form a scheduling task queue; and controlling the palletizer to perform corresponding actions based on the scheduling task queue via the controller's function control module to enable the scheduling and transfer of components on the mold pallet between the curing kiln system and the assembly line system. Specifically, the first status information received from the curing kiln system specifically includes: curing status information of components in the curing kiln, component attribute information, component priority, and curing kiln storage information; and the second status information received from the assembly line system specifically includes: attribute information and component priority of components on the assembly line. In particular, during the palletizing process, the controller is further configured to perform image recognition on the surface state image of the die plate component to obtain a recognition result, and the recognition result is used to indicate whether the die plate component has defects. Figure 3 As shown, the controller 330 includes: an image processing unit 331, which is used to perform grayscale processing on the mold platform component surface state image to obtain a mold platform component surface state grayscale image; a mold platform component surface state semantic coding unit 332, which is used to input the mold platform component surface state grayscale image into a dynamic convolution labeling model including a CNN layer and a Transformer model to obtain a set of mold platform component surface state semantic coding features; an adaptive feature sparsification unit 333, which is used to perform adaptive feature sparsification processing on the set of mold platform component surface state semantic coding features to obtain a set of mold platform component surface state semantic sparse features; a defect recognition unit 344, which is used to determine whether the mold platform component has defects based on the set of mold platform component surface state semantic sparse features.

[0027] Specifically, the image processing unit 331 is configured to grayscale the mold base component surface state image to obtain a grayscale image of the mold base component surface state. It should be understood that when the mold base component is in transit between the curing kiln and the assembly line, the surface state image captured by the industrial camera is converted to a color image. Although color images contain rich color information, color changes are often not a decisive factor for defect detection; instead, changes in brightness and contrast better reflect the geometric form of the mold base component and the texture characteristics of the component surface. Therefore, in the technical solution of the present application, the collected mold base component surface state image is first converted to grayscale to eliminate interference caused by color differences in mold base component recognition, allowing the system to focus more on the component's edges, contours, and other structural features. This is crucial for improving the accuracy and robustness of subsequent image recognition algorithms, especially in environments with varying lighting conditions, complex backgrounds, and diverse mold base component material properties.

[0028] Specifically, the mold base component surface state semantic encoding unit 332 is used to input the mold base component surface state grayscale image into a dynamic convolutional labeling model including a CNN layer and a Transformer model to obtain a set of semantic encoding features of the mold base component surface state. It should be understood that in the field of industrial automation, especially for products such as mold base components with complex shapes and high precision requirements, accurate detection of their surface state is crucial. Traditional pixel-level image recognition methods are difficult to fully adapt to the diversity of mold base component material properties, colors and reflection characteristics, which may lead to misjudgment. The DynamicViT model is a new visual Transformer model that combines convolutional neural networks (CNN) and Transformer models, which can significantly improve the depth and accuracy of understanding the surface state of mold base components. Specifically, the CNN layer can efficiently capture the subtle structure and texture features of the mold base component surface, which is crucial for identifying surface defects of the mold base component; while the Transformer model effectively models long-range dependencies through the self-attention mechanism, which can better process global context information and help the system understand the state of the entire mold base component. This combination enables the system to maintain high-precision defect detection capabilities in the face of changing lighting conditions, complex backgrounds, and diverse material properties of mold base components. In a specific example of the present application, the grayscale image of the mold base component surface state is input into a dynamic convolutional labeling model comprising a CNN layer and a Transformer model to obtain a set of semantic encoding features of the mold base component surface state. Specifically, first, the grayscale image of the mold base component surface state is segmented to obtain multiple local image blocks of the mold base component surface state; then, the CNN layer is used to extract the local features of the mold base component surface state from each local image block to generate multiple local feature maps of the mold base component surface state. Then, all local feature maps are spliced ​​together to form a complete mold base component surface state feature map. This step ensures that information acquired from different perspectives can be integrated to provide a complete perspective for global analysis. Finally, the spliced ​​surface state feature map of the mold base component is input into the Transformer model. At this stage, the model not only considers the importance of local features but also learns from global information to obtain a set of semantically encoded feature vectors for the mold base component's surface state. This set of semantically encoded feature vectors not only reflects the specific morphology of the mold base component's surface but also contains key information about its potential defect patterns.

[0029] Specifically, the adaptive feature sparsification unit 333 is configured to perform adaptive feature sparsification on the set of semantically encoded features of the mold base component surface state to obtain a set of semantically sparse features of the mold base component surface state. Because the set of semantically encoded features of the mold base component surface state, after processing by the CNN layer and the Transformer model, is high-dimensional and contains a lot of redundant information, these feature vectors contain rich information about the mold base component surface state, such as texture, color, and shape. However, they may also contain a large amount of irrelevant or weakly relevant details, which not only increases the computational burden but can also lead to overfitting, affecting the model's generalization and interpretability. Therefore, in order to extract the key features that best represent the surface state of the mold base component from the set of semantic coding features of the mold base component surface state, and focus on those information that are truly useful for detecting defects in the mold base component surface state, the set of semantic coding features of the mold base component surface state is further subjected to adaptive feature sparse processing to obtain a set of semantic sparse features of the mold base component surface state. Through adaptive feature sparse processing, the model can dynamically identify and encode the key points or areas (i.e., "feature distribution hubs") of the distribution of mold base component surface state features in the set of semantic coding features of the mold base component surface state, so as to achieve effective sparse distribution of the surface state feature of the mold base component. In this way, the system can more accurately understand the essential characteristics of the surface state of the mold base component, which can not only better support the subsequent quality control process, but also provide strong support for the subsequent mold base component defect identification, accelerate the discovery and processing of abnormal situations, and thus ensure the safe and stable operation of the production line.

[0030] Specifically, first, the template component surface state semantic feature distribution hub coding feature of the set of template component surface state semantic coding features is calculated. That is, in the technical solution of the present application, the set of template component surface state semantic coding feature vectors is input into the feature distribution hub search network to screen out those parts that best reflect the core information of the template component surface state, so as to obtain the template component surface state semantic feature distribution hub coding vector. In one example, the feature distribution hub search network is used to extract the feature distribution hub coding feature of the set of template component surface state semantic coding feature vectors using the following feature distribution hub calculation formula to obtain the template component surface state semantic feature distribution hub coding vector; wherein, the feature distribution hub calculation formula is:

[0031]

[0032] Wherein, X represents the set of semantic coding feature vectors of the surface state of the mold platform component, x1, x2, x i 、x j and x nare the first, second, i-th, j-th and n-th semantic coding feature vectors of the surface state of the super mold platform component in the set of semantic coding feature vectors of the surface state of the mold platform component, S -1 represents the inverse matrix of the covariance matrix, (·) T is the transpose of the vector, N is the number of vectors in the set of semantic encoding feature vectors of the surface state of the mold platform component minus one, e i represents the semantic weight coefficient of the surface state of the ith mold platform component in the set of semantic weight coefficients of the surface state of the mold platform component, exp represents the exponential function value with the natural constant e as the base, n represents the number of weight coefficients in the set of semantic weight coefficients of the surface state of the mold platform component, v h A hub encoding vector representing the semantic feature distribution of the surface state of the mold platform component.

[0033] Then, based on the module component surface state semantic feature distribution hub coding feature, the set of module component surface state semantic coding features is semantically fine-grained sparsified to obtain the set of module component surface state semantic sparse features. That is, in the technical solution of the present application, first, based on the module component surface state semantic feature distribution hub coding vector, the sparse semantic fine-grained ablation factors of each module component surface state semantic coding feature vector in the set of module component surface state semantic coding feature vectors are calculated respectively to obtain a set of sparse module component surface state semantic fine-grained ablation factors. Specifically, the semantic fine-grained ablation factors of each module component surface state semantic coding feature vector in the set of module component surface state semantic coding feature vectors relative to the module component surface state semantic feature distribution hub coding vector are calculated respectively, so as to measure the relationship between each module component surface state semantic coding feature vector and the module component surface state semantic feature distribution hub coding vector, thereby enhancing the model's ability to distinguish important features. That is, the semantic fine-grained ablation factor is used to quantify the correlation or difference between the semantic encoding feature vector reflecting the surface state of each mold platform component and the hub encoding vector of the semantic feature distribution of the mold platform component surface state. This helps the model understand which features are crucial for describing the surface state of the mold platform component and which features may be redundant or insignificant, thereby improving the accuracy of identifying defects on the surface state of the mold platform component. Next, each mold platform component surface state semantic fine-grained ablation factor in the set of mold platform component surface state semantic fine-grained ablation factors is input into a sparse module based on a gating function to obtain the set of sparse mold platform component surface state semantic fine-grained ablation factors. It should be understood that, because each of the fine-grained semantic ablation factors for the surface state of the template component has different importance in representing the surface state features of the template component, not every fine-grained semantic ablation factor for the surface state of the template component is important for understanding the surface state of the template component. Therefore, each of the fine-grained semantic ablation factors for the surface state of the template component is input into a sparsification module based on a gating function to determine which features should be retained and which should be suppressed, thereby achieving effective sparsification of the distribution of the surface state features of the template component. The gating function dynamically adjusts parameters based on the input set of fine-grained semantic ablation factors for the surface state of the template component, thereby selecting the surface state features of the template component that are helpful for identifying the surface state features of the template component. This makes the sparsified surface state features of the template component more interpretable and expressive in expressing the state of the template component. In this way, the model can focus on those features that best reflect the core information of the surface state of the template component while reducing the influence of redundant information.In one example, the sparse semantic fine-grained ablation factor of each mold platform component surface state semantic coding feature vector in the set of the mold platform component surface state semantic coding feature vector is calculated respectively using the following sparsification factor calculation formula to obtain a set of sparse mold platform component surface state semantic fine-grained ablation factors; wherein the sparsification factor calculation formula is:

[0034]

[0035] Among them, ||·|| is the length of the vector, ε is the adjustment parameter, max is the maximum value, a i represents the semantic fine-grained ablation factor of the surface state of the mold platform component in the set of semantic fine-grained ablation factors of the surface state of the mold platform component, mask is a masking operation, θ is a preset threshold, and w i Represents the i-th sparse module component surface state semantic fine-grained ablation factor in the set of sparse module component surface state semantic fine-grained ablation factors.

[0036] Furthermore, each sparse mold base component surface state semantic fine-grained ablation factor in the set of the sparse mold base component surface state semantic fine-grained ablation factors is used as a weight to weight the set of the mold base component surface state semantic encoding feature vectors to obtain the set of the mold base component surface state semantic sparse feature vectors. That is, by mapping the importance of the features to a numerical range, those features that are considered more valuable for understanding the mold base component surface state are given higher weights, while less important features are given lower weights, so that the mold base component surface state semantic sparse features have better interpretability and higher expressiveness, improve the accuracy of mold base component surface state recognition, and provide strong support for subsequent mold base component defect recognition. In one example, the set of the mold base component surface state semantic encoding feature vectors is weighted by the following weighting formula to obtain the set of the mold base component surface state semantic sparse feature vectors; wherein, the weighting formula is:

[0037] X'={x i '}={w i *x i}

[0038] Among them, x i ' represents the i-th semantic sparse feature vector of the surface state of the mold base component in the set of semantic sparse feature vectors of the surface state of the mold base component, and X' represents the set of semantic sparse feature vectors of the surface state of the mold base component.

[0039] Specifically, the defect recognition unit 334 is used to determine whether the mold base component has defects based on the set of semantic sparse features of the mold base component surface state. That is, in the technical solution of the present application, a linear output module based on a fully connected layer is used to classify the set of semantic sparse features of the mold base component surface state to obtain a classification result, and the classification result is used to indicate whether the mold base component has defects. Here, the linear output module based on a fully connected layer maps the set of semantic sparse features of the mold base component surface state to a specific category, thereby achieving a binary classification of the mold base component surface state: whether there is a defect or not. Specifically, first, the set of semantic sparse features of the mold base component surface state is fully connected encoded using multiple fully connected layers of the classifier to obtain an encoded classification feature vector; and the encoded classification feature vector is passed through a Softmax classification function to obtain the classification result. In this way, accurate recognition of the surface state of the mold base component is achieved, and strong support is provided for efficient production and quality control in the field of industrial automation.

[0040] It should be noted that in the technical solution of the present application, although the feature screening based on the feature distribution hub of the set of semantic coding feature vectors of the surface state of the mold platform component can extract representative features from it to reduce the data processing volume and avoid information redundancy, but since the set of semantic coding feature vectors of the surface state of the mold platform component will inevitably introduce noise during the acquisition process, this will cause the anchoring of the feature distribution hub to have a fine-grained offset and cause selection mismatch in the feature selection process, which will cause the global mold platform component surface state semantic feature representation vector obtained by fully connected encoding to have insufficient long-distance coding representation, thereby reducing the expression effect of the global mold platform component surface state semantic feature representation vector and affecting the accuracy of the attribute label of the first descriptive data item obtained by its input classifier.

[0041] Therefore, in one example, when the global mold platform component surface state semantic feature representation vector is input into a classifier, the global mold platform component surface state semantic feature representation vector is optimized, and the optimization includes the following steps:

[0042] Arranging the eigenvalues ​​of the global mold platform component surface state semantic feature representation vector in ascending order to form a global mold platform component surface state semantic feature representation sequential coding vector;

[0043] In response to the absolute value of the difference between the i-th eigenvalue and the i+1-th eigenvalue of the sequential encoding vector representing the surface state semantic feature of the global mold platform component being less than or equal to the distance difference hyperparameter ε, the weighted sum between the i-th eigenvalue and the i+1-th eigenvalue is calculated as the optimized i+1-th eigenvalue v' i+1 =α×v i+β×v i+1 ;

[0044] In response to the absolute value of the difference between the i-th eigenvalue and the i+1-th eigenvalue of the sequential encoding vector representing the surface state semantic feature of the global mold platform component being greater than the distance difference hyperparameter ε:

[0045] Calculating the square root of the sum of squares of all eigenvalues ​​of the global mold platform component surface state semantic feature representation vector, multiplying the square root by 2 and then dividing it by the square of the length of the global mold platform component surface state semantic feature representation vector to obtain a global mold platform component surface state semantic feature representation space primitive value γ;

[0046] After multiplying the spatial primitive value γ of the global mold platform component surface state semantic feature representation by the i-th eigenvalue, the weighted subtraction between the product and the i+1-th eigenvalue is calculated to obtain the optimized i+1-th eigenvalue v' i+1 =δ×γ×v i -θ×v i+1 ;

[0047] The optimized i+1th eigenvalue of the global template component surface state semantic feature representation sequential coding vector is combined to obtain an optimized global template component surface state semantic feature representation vector, wherein the first eigenvalue of the global template component surface state semantic feature representation sequential coding vector remains unchanged.

[0048] In this way, in order to address the problem of insufficient global fully connected coding representation capability of the feature set of the global template component surface state semantic feature representation vector due to the long distance exceeding the predetermined local distribution interval threshold under the predetermined eigenvalue sequential distribution, the high-dimensional feature space primitive representation of the global template component surface state semantic feature representation vector based on self-inner product fusion is used to capture the complex structure of the global network interaction of its eigenvalues, thereby reconstructing the cross-domain fine-grained aggregation relationship between the eigenvalues ​​of the global template component surface state semantic feature representation vector by simulating the scale-based high-dimensional feature space potential primitives, so as to realize the coding reconstruction of the real serial distribution behavior of the global template component surface state semantic feature representation vector under long distance, improve the coding expression effect of the global template component surface state semantic feature representation vector, and improve the accuracy of the classification result obtained by its input classifier.

[0049] As described above, the palletizer control system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server equipped with a palletizer control algorithm. In one possible implementation, the palletizer control system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the palletizer control system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal. Of course, the palletizer control system 300 can also be one of the many hardware modules of the wireless terminal.

[0050] Alternatively, in another example, the palletizer control system 300 and the wireless terminal may be separate devices, and the palletizer control system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in a predetermined data format.

[0051] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A palletizer control system comprising: Controller, communication module; The controller communicates with the curing kiln system and the assembly line system through the communication module; The controller is further configured to obtain task information of the curing kiln system and the assembly line system through the communication module, and control the action of the palletizer according to the task information to realize the flow of the mold pallet components between the curing kiln and the assembly line, characterized in that it also includes: an industrial camera for capturing an image of a surface state of the die plate component, wherein the die plate component is in a state of being transferred between the curing kiln and the assembly line; The communication module is further used to transmit the surface state image of the die table component to the controller; The controller is further configured to perform image recognition on the surface state image of the die table component to obtain a recognition result, wherein the recognition result is used to indicate whether the die table component has a defect; The controller is further configured to generate a palletizing abnormality warning prompt in response to the recognition result that the mold platform component has a defect; Wherein, the controller includes: An image processing unit, configured to perform grayscale processing on the mold base component surface state image to obtain a mold base component surface state grayscale image; A mold platform component surface state semantic encoding unit, configured to input the mold platform component surface state grayscale image into a dynamic convolutional labeling model comprising a CNN layer and a Transformer model to obtain a set of semantic encoding features of the mold platform component surface state; An adaptive feature thinning unit, configured to perform adaptive feature thinning processing on the set of semantic coding features of the mold platform component surface state to obtain a set of semantic thinning features of the mold platform component surface state; a defect recognition unit, configured to determine whether the mold base component has defects based on a set of semantically sparse features of the surface state of the mold base component; Wherein, the adaptive feature sparsification unit includes: A feature distribution hub coding subunit is used to calculate the template component surface state semantic feature distribution hub coding feature of the set of template component surface state semantic coding features; The fine-grained sparsification subunit is used to perform semantic fine-grained sparsification on the set of semantic coding features of the mold base component surface state based on the distribution hub coding features of the mold base component surface state semantic features to obtain the set of semantic sparse features of the mold base component surface state.

2. The palletizer control system according to claim 1, characterized in that: The mold platform component surface state semantic coding unit is used to: Segmenting the grayscale image of the surface state of the mold platform component to obtain a plurality of local image blocks of the surface state of the mold platform component; Using the CNN layer to extract local features of the mold platform component surface state from each of the multiple mold platform component surface state local image blocks to obtain multiple mold platform component surface state local feature maps; splicing the plurality of local characteristic maps of the surface state of the die table component to obtain a characteristic map of the surface state of the die table component; The mold base component surface state feature map is input into a Transformer model to obtain a set of mold base component surface state semantic coding feature vectors as a set of mold base component surface state semantic coding features.

3. The palletizer control system according to claim 2, characterized in that: The feature distribution hub encoding subunit is used to: The set of the mold base component surface state semantic coding feature vectors is input into a feature distribution hub search network to obtain a mold base component surface state semantic feature distribution hub coding vector as the mold base component surface state semantic feature distribution hub coding feature.

4. The palletizer control system according to claim 3, characterized in that: The fine-grained sparsification subunit includes: A secondary subunit for calculating a sparse factor is used to calculate, based on the module component surface state semantic feature distribution hub coding vector, the sparse semantic fine-grained ablation factor of each module component surface state semantic coding feature vector in the set of module component surface state semantic coding feature vectors to obtain a set of sparse module component surface state semantic fine-grained ablation factors; The feature sparsification secondary subunit is used to perform feature sparsification on the set of semantic encoding feature vectors of the surface state of the mold platform component based on the set of sparse mold platform component surface state semantic fine-grained ablation factors to obtain a set of mold platform component surface state semantic sparse feature vectors as the set of mold platform component surface state semantic sparse features.

5. The palletizer control system according to claim 4, characterized in that: The sparsification factor calculation secondary subunit is used to: Calculating the semantic fine-grained ablation factor of each module platform component surface state semantic coding feature vector in the set of module platform component surface state semantic coding feature vectors relative to the module platform component surface state semantic feature distribution hub coding vector to obtain a set of module platform component surface state semantic fine-grained ablation factors; Each module platform component surface state semantic fine-grained ablation factor in the set of module platform component surface state semantic fine-grained ablation factors is input into a sparse module based on a gating function to obtain the set of sparse module platform component surface state semantic fine-grained ablation factors.

6. The palletizer control system according to claim 5, characterized in that: The feature sparsification secondary subunit is used to: Using each sparse module component surface state semantic fine-grained ablation factor in the set of the sparse module component surface state semantic fine-grained ablation factors as a weight, the set of the module component surface state semantic encoding feature vectors is weighted respectively to obtain the set of the module component surface state semantic sparse feature vectors.

7. The palletizer control system according to claim 6, characterized in that: The defect recognition unit is used to: A linear output module based on a fully connected layer is used to classify the set of semantic sparse features of the surface state of the mold platform component to obtain a classification result, and the classification result is used to indicate whether the mold platform component has defects.

Images