Toy parts equal-portion automatic packaging method and system thereof

Abstract

The present application relates to a kind of toy parts equal-portion quantitative automatic packaging method and system thereof, comprising the following steps: each toy part is sent into conveying belt in the way of scattered flat by the vibration of vibrating disk;Toy parts on conveying belt are photographed using CCD camera;The obtained part image is preprocessed;The image after preprocessing is identified by mechanical vision module;The identified part type is classified;Classification signal is generated based on the classification result;Mechanical arm module is controlled to grasp toy parts according to target category, and it is put into quantitative tray;When the weight reaches the set standard, the parts in the quantitative tray are emptied, and packaged and sent out;The toy parts equal-portion quantitative automatic packaging method and system thereof effectively improve the accuracy, efficiency and flexibility of automatic packaging, solve many technical defects in traditional toy parts packaging method.

Description

Technical Field

[0001] This invention belongs to the field of classification and packaging technology, specifically relating to an automatic method and system for quantitative packaging of toy parts in equal portions. Background Technology

[0002] In the toy manufacturing industry, the automated packaging process for parts is crucial, directly impacting production efficiency and product quality. Currently widely used toy parts packaging technologies suffer from the following major drawbacks:

[0003] Excessive manual intervention: Existing packaging systems often rely on manual sorting, weighing, and packaging of parts, resulting in low production efficiency. This manual intervention not only increases labor intensity but also easily introduces human error, leading to inconsistencies in the quantity and type of packaging, thereby affecting the overall quality of the product.

[0004] Insufficient identification accuracy: Many traditional automated packaging systems use simple sensors or mechanical devices to identify and classify parts. Due to technological limitations, these systems are often inaccurate in identifying part types, easily leading to classification errors and causing chaos and rework in subsequent packaging stages.

[0005] Lack of real-time monitoring and feedback: Existing systems lack real-time monitoring capabilities during parts weighing and sorting, often only conducting quality checks after the entire packaging process is complete. This delayed feedback mechanism makes it difficult to detect problems in a timely manner, affecting packaging consistency and accuracy, and increasing the risk of non-conforming products.

[0006] Low production efficiency: Existing packaging processes are often limited by equipment operating speed and manual operation efficiency, making it impossible to achieve efficient and continuous production. This restricts overall production efficiency and fails to meet rapidly growing market demand.

[0007] In summary, existing toy parts packaging technologies have significant shortcomings in terms of efficiency, accuracy, flexibility, and resource utilization. These shortcomings not only affect enterprises' production efficiency and product quality but also hinder their development in a highly competitive market. Therefore, there is an urgent need for a new automated packaging method to address these issues and improve the efficiency and quality of toy parts packaging. Summary of the Invention

[0008] In order to overcome the shortcomings and disadvantages of the prior art, the primary objective of this invention is to provide an automatic packaging method and system for equal-quantity packaging of toy parts. This automatic packaging method and system for equal-quantity packaging of toy parts effectively improves the accuracy, efficiency and flexibility of automatic packaging, and solves many technical defects in traditional toy parts packaging methods.

[0009] The primary objective of this invention is achieved through the following technical solution:

[0010] An automated method for quantitatively packaging toy parts includes the following steps:

[0011] Toy parts are poured into a vibratory feeder, and the vibration of the feeder sends the toy parts into the conveyor belt in a dispersed and flat manner.

[0012] A CCD camera was used to photograph toy parts on a conveyor belt to obtain images of the parts.

[0013] Preprocess the acquired part images;

[0014] A machine vision module is used to identify the part type from the pre-processed image.

[0015] Classify the identified part types;

[0016] A classification signal is generated based on the recognition and classification results to represent the category information of each part;

[0017] Upon receiving the classification signal, the robotic arm module is controlled to grasp toy parts according to the target category and place them into a quantitative tray;

[0018] The pressure sensor under the quantitative pan is used for real-time weighing. When the weight reaches the set standard, the parts in the quantitative pan are emptied, packaged, and sent out.

[0019] As a preferred method, the toy parts on the conveyor belt are photographed using a CCD camera to obtain images of the parts.

[0020] An adjustable-brightness LED lighting device is installed on one side of the CCD camera to take pictures of toy parts on the conveyor belt;

[0021] A photoelectric sensor is installed at the feed inlet of the conveyor belt. When a part passes through the photoelectric sensor, it triggers a CCD camera to acquire an image.

[0022] As a preferred method, the preprocessing method for the acquired part images is as follows:

[0023] Adjust the image to a uniform 224. 224 pixels;

[0024] Normalize the image pixel values ​​to the range [0, 1];

[0025] Data augmentation techniques are used to enhance the image data.

[0026] As a preferred option, a machine vision module is used to identify the part type of the pre-processed image;

[0027] The preprocessed part images are input into a convolutional neural network to extract feature information;

[0028] The extracted features are nonlinearly transformed using a multi-layered convolutional neural network to generate the category probability distribution for each part.

[0029] Based on the generated category probability distribution, the category with the highest probability is selected as the identification result of the part;

[0030] The recognition results are converted into classification signals, which are used to control the robotic arm module to grasp the corresponding toy parts.

[0031] As a preferred method, the construction method of the convolutional neural network model is as follows:

[0032] Construct at least one convolutional layer, wherein the convolutional layer uses a 5x5 kernel, a stride of 1, and a ReLU activation function;

[0033] Construct at least one pooling layer, which is a max pooling layer with a window size of 2x2 and a stride of 2;

[0034] Construct at least one fully connected layer whose output unit count equals the number of categories, and use the softmax activation function to output the classification probability;

[0035] The Adam optimizer was chosen as the training algorithm, and the initial learning rate was set to 0.001.

[0036] Cross-entropy is used as the loss function to evaluate the classification performance of the model;

[0037] Use early stopping to monitor validation loss and prevent overfitting;

[0038] The model performance was evaluated on a separate test set, and its accuracy, precision, recall, and F1-score were calculated.

[0039] A confusion matrix was plotted to analyze the recognition performance of different part categories;

[0040] The model was adjusted based on the evaluation results and retrained to further improve classification performance.

[0041] As a preferred method, the method for classifying the identified part types is as follows:

[0042] The preprocessed part images are input into a convolutional neural network to extract feature information;

[0043] By using the multi-layer structure of CNN, the extracted features are subjected to non-linear transformation to generate the category probability distribution of each part;

[0044] Based on the generated category probability distribution, the category with the highest probability is selected as the identification result for the part;

[0045] The recognition results are converted into classification signals.

[0046] Preferably, the method of receiving the classification signal and controlling the robotic arm module to grasp toy parts according to the target category and place them into a quantitative tray is as follows:

[0047] The robotic arm control system receives classification signals from the machine vision module via a communication interface. These signals contain the identified part types and their corresponding position information.

[0048] Analyze the information in the signal to extract the category and location information of the target part;

[0049] Based on the parsed location information, determine the specific location of each type of part on the conveyor belt;

[0050] The robotic arm control system calculates the optimal grasping path from the current position to the target part;

[0051] Based on the path planning results, a motion command is issued to control the robotic arm to move above the target part;

[0052] Once the robotic arm reaches the target location, it opens its gripper to perform the grasping action.

[0053] After grasping the part, the robotic arm moves to the top of the quantitative tray according to the preset path;

[0054] Open the gripper to release the part into the metering pan;

[0055] The robotic arm continues to receive classification signals and perform cyclical operations, constantly grabbing different types of parts and placing them into a quantitative tray until the quantitative requirements are met.

[0056] Another technical problem to be solved by the present invention is to provide an automatic packaging system for equal-quantity packaging of toy parts, comprising:

[0057] Vibratory feeder module is used to receive and disperse toy parts, feeding them into the conveyor belt in a flat manner;

[0058] The image acquisition module includes a CCD camera, configured to capture images of toy parts on a conveyor belt and obtain images of the parts;

[0059] The image processing module is used to preprocess the image of the part;

[0060] The machine vision module is used to identify the part type from pre-processed part images;

[0061] The classification module is used to classify the identified part types and generate classification signals representing information about each part category.

[0062] The control system module receives classification signals and controls the robotic arm module.

[0063] The robotic arm module, according to the instructions of the control system module, grabs toy parts of a specified category and places them into a quantitative tray;

[0064] A quantitative tray module is used to store toy parts dispensed by a robotic arm;

[0065] A weighing sensor module is installed below the quantitative pan to perform real-time weighing to monitor the weight of the parts;

[0066] The packaging module is used to empty the parts in the quantitative tray and package them after the weight reaches the set standard.

[0067] Another technical problem to be solved by the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the automatic packaging method for equal-quantity toy parts as described above.

[0068] Another technical problem to be solved by the present invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the automatic packaging method for equal-quantity toy parts as described above.

[0069] Compared with the prior art, the present invention has the following advantages:

[0070] This solution enables continuous and efficient parts sorting and packaging through automated vibratory feeders and robotic arms, significantly improving overall production efficiency. Manual operation often leads to errors, such as inaccurate parts quantity or misclassification. Using CCD cameras and mechanical vision modules for parts identification and classification can effectively reduce human error and improve the accuracy of parts classification.

[0071] This solution utilizes image preprocessing and a machine vision module to ensure accurate identification and classification of various types of parts, avoiding confusion and classification errors. Real-time weighing using pressure sensors ensures that parts in the quantitative pan reach the set standard before being emptied and packaged. This real-time monitoring mechanism allows the system to quickly provide feedback and adjust, ensuring the accuracy of each packaging and reducing the generation of defective products.

[0072] This automated packaging method can generate classification signals based on the identified part categories, controlling the robotic arm to grasp the parts. This means the system can adapt to different types and quantities of parts, effectively meeting the market's demand for diversified products. By reducing human intervention, improving production efficiency, and reducing material waste, this method can effectively reduce production costs in the long run, thereby enhancing the company's competitiveness.

[0073] By optimizing automated packaging systems, waste of materials and human resources is reduced, which meets the requirements of sustainable development. Automated quantitative packaging methods ensure that the quantity and quality of parts in each package are consistent, thereby improving the overall quality of the final product and enhancing customer satisfaction. Attached Figure Description

[0074] Figure 1 This is a flowchart of an automatic packaging method for equal-quantity packaging of toy parts according to the present invention. Detailed Implementation

[0075] The specific embodiments of the present invention will be further described in detail to make the technical solution of the present invention easier to understand and master.

[0076] Example

[0077] See Figure 1 As shown, an automatic method for quantitatively packaging toy parts includes the following steps:

[0078] Toy parts are poured into a vibratory feeder, and the vibration of the feeder sends the toy parts into the conveyor belt in a dispersed and flat manner.

[0079] A CCD camera was used to photograph toy parts on a conveyor belt to obtain images of the parts.

[0080] Preprocess the acquired part images;

[0081] A machine vision module is used to identify the part type from the pre-processed image.

[0082] Classify the identified part types;

[0083] A classification signal is generated based on the recognition and classification results to represent the category information of each part;

[0084] Upon receiving the classification signal, the robotic arm module is controlled to grasp toy parts according to the target category and place them into a quantitative tray;

[0085] The pressure sensor under the quantitative pan is used for real-time weighing. When the weight reaches the set standard, the parts in the quantitative pan are emptied, packaged, and sent out.

[0086] Automated vibratory feeders, conveyor belts, and robotic arms reduce manual intervention time, increase overall production speed, and enable 24-hour uninterrupted production. Real-time weighing using pressure sensors ensures accurate quantity of parts for each package, reducing human error and improving packaging accuracy. The automation system reduces reliance on manual labor, lowering labor costs, especially in large-scale production, where it effectively reduces labor expenses.

[0087] The automated parts identification and classification process reduces quality fluctuations caused by human factors, ensuring consistent parts quality in each package and thus improving the overall quality of the final product. The system can generate classification signals based on the identified parts type, flexibly handling different types and sizes of toy parts and adapting to the market's demand for diversified products.

[0088] By monitoring the weight and classification status of parts in real time, the system can quickly provide feedback and adjust the workflow, ensuring the smooth operation of the production line and reducing downtime. By using robotic arms and automated equipment, potential safety hazards from manual operation are reduced, and the safety of the working environment is improved. The automated quantitative control system ensures the accuracy of the number of parts, avoids material waste due to over- or under-quantity, and improves resource utilization efficiency.

[0089] An efficient packaging system enables enterprises to respond more quickly to market changes and customer needs, thereby gaining a competitive advantage. The system can be upgraded and expanded as needed to support new parts types and packaging requirements, demonstrating good sustainability. Through automation, real-time collection and analysis of production data can be achieved, providing data support for subsequent production optimization. It reduces manual intervention and material waste, helping to reduce the environmental impact of production and meeting the requirements of modern enterprises for sustainable development.

[0090] The method of acquiring images of toy parts by photographing them on a conveyor belt using a CCD camera is as follows:

[0091] An adjustable-brightness LED lighting device is installed on one side of the CCD camera to take pictures of toy parts on the conveyor belt;

[0092] A photoelectric sensor is installed at the feed inlet of the conveyor belt. When a part passes through the photoelectric sensor, it triggers a CCD camera to acquire an image.

[0093] The adjustable brightness of the LED lighting device can optimize the lighting effect according to different ambient lighting conditions and component materials, thereby improving image clarity and contrast and ensuring image quality. The CCD camera is triggered by a photoelectric sensor to capture images, ensuring that the image is captured the instant the component passes by, avoiding missed or false images and improving the efficiency and accuracy of image acquisition. The automated design of this solution reduces manual intervention, improves the intelligence level of the entire system, and reduces the workload of operators.

[0094] Photoelectric sensors can quickly respond to the passing of parts, and CCD cameras capture images almost instantly as the parts pass by, reducing acquisition latency and making them suitable for high-speed production lines. Adjustable LED lighting can be adjusted according to the color, shape, and surface characteristics of different parts, ensuring that different types of parts can achieve the best shooting results in various environments. High-quality images help subsequent machine vision modules to more accurately identify part types, reduce the recognition error rate, and improve the overall packaging accuracy.

[0095] Compared to traditional light sources, LED lighting has lower energy consumption and a longer lifespan, reducing energy costs and maintenance frequency in the production process. The brightness of LED lighting and the setting of photoelectric sensors can be adjusted according to production needs, increasing the system's flexibility and adapting to different production requirements. The combination of LED lighting and photoelectric sensors reduces the noise of mechanical equipment and improves the comfort of the working environment.

[0096] The method is simple in design, making later maintenance and upgrades more convenient, and technicians can easily make adjustments and troubleshoot problems; high-quality images captured by the CCD camera can be stored and analyzed, providing data support for subsequent quality control and production optimization; the system's automated design reduces direct contact between operators and machines, lowering potential safety risks.

[0097] The method for preprocessing the acquired part images is as follows:

[0098] Adjust the image to a uniform 224. 224 pixels;

[0099] Normalize the image pixel values ​​to the range [0, 1];

[0100] Data augmentation techniques are used to enhance the image data.

[0101] Resizing images to a uniform 224×224 pixels ensures consistent input size, facilitating training or inference in neural networks. This is a standard input requirement for many deep learning models, reducing the complexity caused by different input sizes. A uniform size also simplifies image processing, avoiding image alignment or padding issues caused by varying dimensions.

[0102] Normalizing image pixel values ​​to the range [0, 1] can improve the convergence speed of the model and reduce the difficulty of training. Normalization can prevent gradient explosion or vanishing, helping the model learn faster; after normalization, the distribution of input data is more consistent, which can reduce the sensitivity to the learning rate and improve the stability and performance of the model.

[0103] Data augmentation techniques (such as rotation, scaling, and flipping) can generate diverse training samples to increase the model's generalization ability. This can effectively improve the model's performance on new data and reduce the risk of overfitting; data augmentation can simulate different shooting conditions, angles, and environmental changes, making the model more robust to real-world applications; data augmentation can create more samples on limited datasets, improving training efficiency and helping the model learn more comprehensive features.

[0104] Increasing the number and diversity of training samples can effectively reduce overfitting of the model on the training set, making it perform better on the validation and test sets; preprocessed images can better reflect the features of parts, enabling the model to identify and classify more accurately, thereby improving overall performance; uniform size and normalization help convolutional neural networks extract image features better, improving the model's recognition ability.

[0105] A machine vision module is used to identify the part type from the pre-processed image.

[0106] The preprocessed part images are input into a convolutional neural network to extract feature information;

[0107] The extracted features are nonlinearly transformed using a multi-layered convolutional neural network to generate the category probability distribution for each part.

[0108] Based on the generated category probability distribution, the category with the highest probability is selected as the identification result of the part;

[0109] The recognition results are converted into classification signals, which are used to control the robotic arm module to grasp the corresponding toy parts.

[0110] The machine vision module enables rapid and automatic identification of parts, reducing manual intervention and improving work efficiency. The system can perform real-time image processing and recognition, making it suitable for industrial applications requiring rapid response, such as automated grasping on production lines. Convolutional neural networks have powerful feature extraction capabilities, automatically identifying and extracting important features from images, reducing the need for manual feature engineering. The nonlinear transformation of the multi-layer structure can effectively capture complex patterns, improving classification accuracy and making it suitable for complex part identification tasks.

[0111] By generating category probability distributions, the classification results of each part can be evaluated more meticulously, providing a more reliable basis for identification. Selecting the category with the highest probability as the identification result helps improve the accuracy of identification and reduce false positives and false negatives. The identification results can be quickly converted into control signals to drive the robotic arm module to perform precise grasping, improving the intelligence level of the system. It can adapt to different part types and environmental changes, realizing flexible grasping strategies.

[0112] This solution employs a machine vision module and a convolutional neural network, which can be easily integrated with other systems to form a complete automated production line. It can expand the types of parts that can be recognized or enhance other functions as needed, such as adding classifications for multiple parts. It reduces errors caused by human operation, ensuring the stability and reliability of the recognition process. The combination of machine vision and CNN can provide consistent recognition standards, improving production quality.

[0113] Automated identification and grasping systems reduce labor costs, improve production efficiency, and save time; they can achieve higher production accuracy and consistency without affecting efficiency, thereby optimizing the entire production process; this solution is not only applicable to toy parts, but can also be extended to other fields, such as the identification and grasping of various industrial products like automotive parts and electronic components; it can be adjusted and optimized according to different product characteristics and application scenarios.

[0114] The method for constructing a convolutional neural network model is as follows:

[0115] Construct at least one convolutional layer, wherein the convolutional layer uses a 5x5 kernel, a stride of 1, and a ReLU activation function;

[0116] Construct at least one pooling layer, which is a max pooling layer with a window size of 2x2 and a stride of 2;

[0117] Construct at least one fully connected layer whose output unit count equals the number of categories, and use the softmax activation function to output the classification probability;

[0118] The Adam optimizer was chosen as the training algorithm, and the initial learning rate was set to 0.001.

[0119] Cross-entropy is used as the loss function to evaluate the classification performance of the model;

[0120] Use early stopping to monitor validation loss and prevent overfitting;

[0121] The model performance was evaluated on a separate test set, and its accuracy, precision, recall, and F1-score were calculated.

[0122] A confusion matrix was plotted to analyze the recognition performance of different part categories;

[0123] The model was adjusted based on the evaluation results and retrained to further improve classification performance.

[0124] Normalization and one-hot encoding were performed using the CIFAR-10 dataset;

[0125] Convolutional layers are constructed using Conv2D, with kernel size set to 5x5, stride to 1, and activation function set to ReLU.

[0126] The pooling layer is constructed using MaxPooling2D with a window size of 2x2 and a stride of 2.

[0127] After flattening the feature map using Flatten, a fully connected layer is constructed, and the final output layer uses the softmax activation function.

[0128] We selected the Adam optimizer with an initial learning rate of 0.001 and used cross-entropy as the loss function.

[0129] Monitor and validate the loss to prevent model overfitting.

[0130] During training, 20% of the training data is used as the validation set;

[0131] Evaluate the model on the test set and output the test loss and accuracy.

[0132] Use `classification_report` to output metrics such as precision, recall, and F1-score.

[0133] Calculate and plot the confusion matrix, and analyze the recognition effect of different part categories.

[0134] Using a 5x5 convolution kernel can capture local features in an image, such as edges and textures. This feature extraction capability enables the model to automatically learn and recognize complex patterns. The ReLU activation function (Rectified LinearUnit) introduces non-linearity, allowing the model to learn more complex features while avoiding the vanishing gradient problem and accelerating model convergence.

[0135] Using a 2x2 max pooling layer effectively reduces the size of the feature maps while preserving important features. This not only reduces computational cost but also improves the model's generalization ability and reduces the risk of overfitting. Fully connected layers and the softmax activation function: The fully connected layer design allows the network to synthesize feature information from the convolutional layers, while the softmax activation function provides a multi-class probability distribution, suitable for outputting multi-class classification problems.

[0136] Adam was chosen as the optimizer because it combines the advantages of momentum and RMSProp, is highly adaptable, accelerates the model training process, and is suitable for most deep learning tasks. The cross-entropy loss function performs well in multi-class classification tasks and can effectively evaluate the model's classification performance. Early stopping by monitoring the validation loss can effectively prevent overfitting and improve the model's generalization ability on the test set.

[0137] Evaluating model performance on independent test sets provides a true reflection of its generalization ability. By calculating accuracy, precision, recall, and F1-score, a comprehensive assessment of the model's performance across different aspects can provide a clear direction for subsequent optimization. The confusion matrix visually demonstrates the model's recognition performance across different categories, helping to analyze the model's weaknesses and guiding further model improvements and training strategy adjustments.

[0138] The method for classifying the identified part types is as follows:

[0139] The preprocessed part images are input into a convolutional neural network to extract feature information;

[0140] By using the multi-layer structure of CNN, the extracted features are subjected to non-linear transformation to generate the category probability distribution of each part;

[0141] Based on the generated category probability distribution, the category with the highest probability is selected as the identification result for the part;

[0142] The recognition results are converted into classification signals.

[0143] Traditional image processing methods rely on manually designed features, while CNNs can automatically extract local features (such as edges, corners, and textures) from part images through convolutional layers. These features are crucial for part recognition and can adapt to different scales and locations. By automating feature extraction, the step of manually designing features is eliminated, making the model more efficient and able to handle more complex situations.

[0144] Convolutional Neural Networks (CNNs) consist of multiple convolutional layers, pooling layers, and fully connected layers. This multi-layered structure allows the model to progressively extract more abstract and complex features. As the number of network layers increases, CNNs can gradually abstract from simple low-level features such as edges and shapes to higher-level part features. The ReLU activation function in each layer enables the model to learn non-linear features, which is crucial for complex part classification tasks. Simple linear models may not be able to effectively distinguish complex part categories, while CNNs can capture more diverse image features through multi-layered non-linear transformations.

[0145] The final output layer of a CNN (such as a softmax layer) generates a probability distribution for each part belonging to each category. This probability output method is more flexible than directly outputting category labels and provides richer information, such as the uncertainty of classification. If needed, different parts can be treated differently based on these probabilities (e.g., setting a threshold to only accept high-probability classifications); the category with the highest probability is selected as the final recognition result. This simple and intuitive decision-making method ensures high classification accuracy in most cases.

[0146] The recognition results of CNNs are directly converted into classification signals, facilitating subsequent system processing. These signals can be directly used for decisions in automated production lines and control of machine vision modules, eliminating the need for complex post-processing steps and reducing latency and complexity. Because CNNs can automatically learn useful features from data, and their hierarchical structure has strong fault tolerance, they can adapt to different types of part images, including different shooting angles, lighting conditions, and background noise. This makes the model more robust and capable of generalization in practical applications; even parts with different shapes, sizes, and textures can be effectively distinguished by CNNs through multi-layer feature extraction and classification probability distribution.

[0147] CNNs can learn effectively on large amounts of labeled training data, thereby capturing complex relationships between part features. Through backpropagation and optimization algorithms (such as Adam), model parameters can be efficiently tuned to continuously improve classification accuracy. Due to the hierarchical structure and efficient parameter sharing of CNNs, model training is usually faster and has better performance than traditional machine learning methods.

[0148] The method for receiving the classification signal and controlling the robotic arm module to grasp toy parts according to the target category and place them into a quantitative tray is as follows:

[0149] The robotic arm control system receives classification signals from the machine vision module via a communication interface. These signals contain the identified part types and their corresponding position information.

[0150] Analyze the information in the signal to extract the category and location information of the target part;

[0151] Based on the parsed location information, determine the specific location of each type of part on the conveyor belt;

[0152] The robotic arm control system calculates the optimal grasping path from the current position to the target part;

[0153] Based on the path planning results, a motion command is issued to control the robotic arm to move above the target part;

[0154] Once the robotic arm reaches the target location, it opens its gripper to perform the grasping action.

[0155] After grasping the part, the robotic arm moves to the top of the quantitative tray according to the preset path;

[0156] Open the gripper to release the part into the metering pan;

[0157] The robotic arm continues to receive classification signals and perform cyclical operations, constantly grabbing different types of parts and placing them into a quantitative tray until the quantitative requirements are met.

[0158] The entire grasping and delivery process is automated, reducing human intervention and improving production efficiency. The robotic arm can operate accurately based on classification signals, saving time on manual inspection and operation; the robotic arm can work continuously, automatically receiving signals and performing repetitive operations, making it suitable for large-scale production needs and increasing production capacity.

[0159] With the position information provided by the machine vision module, the robotic arm can accurately locate the target part, reducing the possibility of grasping errors; through cyclic operation, the robotic arm can ensure that each part is grasped and placed, reducing the risk of omissions and errors.

[0160] This system can handle various types of toy parts and can adapt to different production needs simply by adjusting the classification signals, demonstrating excellent flexibility. The system can receive and process new classification signals in real time, enabling it to adapt to changes in parts on the production line and ensure a smooth production process.

[0161] The robotic arm can calculate the optimal grasping path from its current position to the target part, reducing travel time and energy consumption and improving work efficiency; through path planning, the system can avoid collisions with other objects, ensuring the safe operation of the robotic arm.

[0162] The robotic arm can quickly receive classification signals and perform grasping and delivery operations, improving the overall efficiency of the production line; the system can set quantitative requirements to ensure that parts are delivered as needed during the production process, avoiding waste or overproduction.

[0163] The robotic arm's status and operation can be monitored in real time through the communication interface, facilitating management and maintenance; the system can be set with an alarm mechanism to provide timely feedback on the robotic arm's working status, enabling rapid diagnosis and repair of faults.

[0164] An automated packaging system for equal-quantity packaging of toy parts includes:

[0165] Vibratory feeder module is used to receive and disperse toy parts, feeding them into the conveyor belt in a flat manner;

[0166] The image acquisition module includes a CCD camera, configured to capture images of toy parts on a conveyor belt and obtain images of the parts;

[0167] The image processing module is used to preprocess the image of the part;

[0168] The machine vision module is used to identify the part type from pre-processed part images;

[0169] The classification module is used to classify the identified part types and generate classification signals representing information about each part category.

[0170] The control system module receives classification signals and controls the robotic arm module.

[0171] The robotic arm module, according to the instructions of the control system module, grabs toy parts of a specified category and places them into a quantitative tray;

[0172] A quantitative tray module is used to store toy parts dispensed by a robotic arm;

[0173] A weighing sensor module is installed below the quantitative pan to perform real-time weighing to monitor the weight of the parts;

[0174] The packaging module is used to empty the parts in the quantitative tray and package them after the weight reaches the set standard.

[0175] The system is highly automated, with all modules working together to reduce manual intervention and improve production efficiency. The combination of vibratory feeder and conveyor belt can quickly disperse and transport parts, reducing waiting and processing time. CCD camera and image processing module can efficiently capture and process part images, ensuring that parts are accurately identified. Mechanical vision module and classification module can intelligently classify parts in real time, ensuring that each type of part can be correctly identified and processed.

[0176] The weighing sensor module monitors the weight of parts in real time, ensuring that the quantity and quality of each package meet the set standards, thus improving product consistency and reliability. The fully automated system reduces human error and ensures the stable quality of each batch of products. The system can handle various types of toy parts and can flexibly adjust the production line configuration according to market demand. In terms of design, modules can be added or replaced according to production needs, such as introducing new recognition algorithms or replacing robotic arms, to improve production capacity.

[0177] The weighing sensor provides real-time feedback, enabling timely monitoring of the weight of parts in the quantitative pan to ensure accurate packaging quantities; the coordinated operation of various modules in the system provides real-time monitoring, timely detection and handling of faults, and improved system reliability; due to the high degree of automation of the system, it can significantly reduce reliance on manual operation and lower labor costs; accurate weighing and sorting reduce material waste and improve resource utilization.

[0178] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the automatic packaging method for equal-quantity toy parts as described above.

[0179] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the automatic packaging method for equal-quantity toy parts as described above.

[0180] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0181] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0182] Of course, the above are merely typical examples of the present invention. Besides these, the present invention can have many other specific embodiments. All technical solutions formed by equivalent substitution or equivalent transformation fall within the scope of protection claimed by the present invention.

Claims

1. A method for automatically packaging toy parts in equal quantities, characterized in that... This includes the following steps: Toy parts are poured into a vibratory feeder, and the vibration of the feeder sends the toy parts into the conveyor belt in a dispersed and flat manner. A CCD camera was used to photograph toy parts on a conveyor belt to obtain images of the parts. Preprocess the acquired part images; A machine vision module is used to identify the part type from the pre-processed image. Classify the identified part types; A classification signal is generated based on the recognition and classification results to represent the category information of each part; Upon receiving the classification signal, the robotic arm module is controlled to grasp toy parts according to the target category and place them into a quantitative tray; The pressure sensor under the quantitative pan is used for real-time weighing. When the weight reaches the set standard, the parts in the quantitative pan are emptied, packaged and sent out. The method for preprocessing the acquired part images is as follows: Adjust the image to a uniform 224. 224 pixels; Normalize the image pixel values ​​to the range [0, 1]; Data augmentation techniques are used to enhance the image data; A machine vision module is used to identify the part type from the pre-processed image. The preprocessed part images are input into a convolutional neural network to extract feature information; The extracted features are nonlinearly transformed using a multi-layered convolutional neural network to generate the category probability distribution for each part. Based on the generated category probability distribution, the category with the highest probability is selected as the identification result of the part; The recognition results are converted into classification signals to control the robotic arm module to grasp the corresponding toy parts; The method for constructing a convolutional neural network model is as follows: Construct at least one convolutional layer, wherein the convolutional layer uses a 5x5 kernel, a stride of 1, and a ReLU activation function; Construct at least one pooling layer, which is a max pooling layer with a window size of 2x2 and a stride of 2; Construct at least one fully connected layer whose output unit count equals the number of categories, and use the softmax activation function to output the classification probability; The Adam optimizer was chosen as the training algorithm, and the initial learning rate was set to 0.

001. Cross-entropy is used as the loss function to evaluate the classification performance of the model; Use early stopping to monitor validation loss and prevent overfitting; The model performance was evaluated on a separate test set, and its accuracy, precision, recall, and F1-score were calculated. A confusion matrix was plotted to analyze the recognition performance of different part categories; The model was adjusted based on the evaluation results and retrained to further improve classification performance.

2. The automatic packaging method for equal-quantity packaging of toy parts according to claim 1, characterized in that, The method of acquiring images of toy parts by photographing them on a conveyor belt using a CCD camera is as follows: An adjustable-brightness LED lighting device is installed on one side of the CCD camera to take pictures of toy parts on the conveyor belt; A photoelectric sensor is installed at the feed inlet of the conveyor belt. When a part passes through the photoelectric sensor, it triggers a CCD camera to acquire an image.

3. The automatic packaging method for equal-quantity packaging of toy parts according to claim 1, characterized in that, The method for classifying the identified part types is as follows: The preprocessed part images are input into a convolutional neural network to extract feature information; By using the multi-layer structure of CNN, the extracted features are subjected to non-linear transformation to generate the category probability distribution of each part; Based on the generated category probability distribution, the category with the highest probability is selected as the identification result for the part; The recognition results are converted into classification signals.

4. The automatic packaging method for equal-quantity packaging of toy parts according to claim 1, characterized in that, The method for receiving the classification signal and controlling the robotic arm module to grasp toy parts according to the target category and place them into a quantitative tray is as follows: The robotic arm control system receives classification signals from the machine vision module via a communication interface. These signals contain the identified part types and their corresponding position information. Analyze the information in the signal to extract the category and location information of the target part; Based on the parsed location information, determine the specific location of each type of part on the conveyor belt; The robotic arm control system calculates the optimal grasping path from the current position to the target part; Based on the path planning results, a motion command is issued to control the robotic arm to move above the target part; Once the robotic arm reaches the target location, it opens its gripper to perform the grasping action. After grasping the part, the robotic arm moves to the top of the quantitative tray according to the preset path; Open the gripper to release the part into the metering pan; The robotic arm continues to receive classification signals and perform cyclical operations, constantly grabbing different types of parts and placing them into a quantitative tray until the quantitative requirements are met.

5. An automated packaging system for equal-quantity packaging of toy parts using the method described in any one of claims 1-4, characterized in that, Including: Vibratory feeder module is used to receive and disperse toy parts, feeding them into the conveyor belt in a flat manner; The image acquisition module includes a CCD camera, configured to capture images of toy parts on a conveyor belt and obtain images of the parts; The image processing module is used to preprocess the image of the part; The machine vision module is used to identify the part type from pre-processed part images; The classification module is used to classify the identified part types and generate classification signals representing information about each part category. The control system module receives classification signals and controls the robotic arm module. The robotic arm module, according to the instructions of the control system module, grabs toy parts of a specified category and places them into a quantitative tray; A quantitative tray module is used to store toy parts dispensed by a robotic arm; A weighing sensor module is installed below the quantitative pan to perform real-time weighing to monitor the weight of the parts; The packaging module is used to empty the parts in the quantitative tray and package them after the weight reaches the set standard.

6. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the automatic packaging method for equal-quantity packaging of toy parts as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the automatic packaging method for equal-quantity packaging of toy parts as described in any one of claims 1-4.

Images