Aluminum foil seal detection method and system based on deep learning

Abstract

The invention discloses an aluminum foil seal detection method and system based on deep learning, and the method comprises the steps: obtaining a large amount of infrared image data of a heated aluminum foil seal, and obtaining the sealing type of the aluminum foil seal; preprocessing the aluminum foil sealing infrared image data, and marking the preprocessed aluminum foil sealing infrared image data by using a corresponding sealing category; inputting the marked aluminum foil seal infrared image data into an aluminum foil seal detection network model for parameter training and testing; and obtaining a to-be-detected aluminum foil seal infrared image, and inputting the to-be-detected aluminum foil seal infrared image into the trained aluminum foil seal detection network model to obtain a seal category. According to the invention, the sealing condition of the aluminum foil seal can be quickly and accurately obtained.

Description

Technical Field

[0001] The present application relates to the field of detection technology, and in particular to an aluminum foil seal detection method and system. Background Art

[0002] Aluminum foil sealing technology is widely used in the pharmaceutical, food and cosmetic industries. Its main functions include ensuring the sealing and anti-counterfeiting properties of products, extending the shelf life and providing a convenient sealing process. Aluminum foil gaskets can effectively block air, moisture and other pollutants, ensure the purity and quality of the product in the bottle, thereby improving the anti-counterfeiting ability of the product and preventing counterfeit and inferior products from entering the market. Aluminum foil sealing can also effectively extend the shelf life of the product and maintain the efficacy and quality of the product. In addition, the sealing process is highly automated, suitable for large-scale production, and significantly improves production efficiency.

[0003] At present, the detection of aluminum foil seals mainly relies on the following three methods:

[0004] (1) Manual visual inspection: Manual visual inspection is performed by an operator who unscrews the bottle cap and visually inspects the appearance of the seal to determine the integrity and quality of the seal. The operation is simple but time-consuming and labor-intensive, with low detection efficiency and high subjectivity, and is prone to missed detection and false detection.

[0005] (2) Mechanical inspection: Mechanical inspection uses a mechanical device to apply pressure to the bottle cap or perform visual inspection to detect the tightness and integrity of the seal. Although it is relatively automated, it is only suitable for detecting simple defects and has difficulty identifying minor or complex sealing problems. The detection accuracy is limited.

[0006] (3) Infrared detection: In recent years, infrared thermal imaging technology has been introduced into the field of seal detection. The aluminum foil gasket is heated by electromagnetic induction, and the temperature distribution image of the bottle cap seal is captured by an infrared camera. However, most of the existing infrared imaging-based detection methods rely on traditional image processing technologies, such as edge detection, threshold segmentation, template matching, etc. These methods have the problems of low detection accuracy and poor robustness when faced with complex temperature distribution and diverse defect forms.

[0007] Therefore, existing detection technologies have certain deficiencies in efficiency and accuracy. Summary of the invention

[0008] Based on this, in response to the above technical problems, a deep learning-based aluminum foil seal detection method and system are provided to solve the problem of low detection efficiency or low accuracy in the prior art.

[0009] In a first aspect, a method for detecting aluminum foil seals based on deep learning is provided, the method comprising:

[0010] Acquire a large amount of infrared image data of the heated aluminum foil seal collected by an infrared thermal imager, and obtain the sealing condition of the aluminum foil seal;

[0011] Acquire a large amount of infrared image data of the heated aluminum foil seal collected by an infrared thermal imager, and obtain the sealing condition of the aluminum foil seal;

[0012] The labeled aluminum foil seal infrared image data is input into the aluminum foil seal detection network model for parameter training and testing;

[0013] An infrared image of the aluminum foil seal to be tested is obtained, and the infrared image of the aluminum foil seal to be tested is input into a trained aluminum foil seal detection network model to obtain a sealing level.

[0014] In the above scheme, optionally, the preprocessing includes: cleaning, denoising, lossless conversion, and data enhancement.

[0015] In the above scheme, optionally, the sealing detection network model includes: an input layer, a backbone network, a path aggregation network, a detection head and a classification optimization module.

[0016] In the above scheme, further optionally, the backbone network adopts a CSPDarknet53 structure, including six feature extraction modules and one SPP module, and the feature extraction module is composed of multiple ConvBlocks and CSPBlock_X;

[0017] The feature extraction module is used to extract primary features, middle-level features, and deep-level features, and perform feature fusion on the primary features, middle-level features, and deep-level features;

[0018] The SPP module includes maximum pooling layer modules of different sizes, and performs spatial pyramid pooling on the feature map extracted by the residual module in the feature extraction module.

[0019] In the above scheme, further optionally, the path aggregation network includes upsampling, downsampling, convolution, feature fusion and jump connection, and multi-scale features are fused through top-down and bottom-up paths.

[0020] In the above solution, further optionally, the detection head comprises three output layers for performing target detection at different scales and outputting the aluminum foil sealing level;

[0021] The output layer includes a convolutional layer, a batch normalization layer, and an activation function.

[0022] In the above scheme, further optionally, the classification optimization module uses loss function optimization, data enhancement and regularization; the loss function optimization adopts Focal LoSs and Cross-Entropy Loss; the regularization uses Dropout and Batch Normalization.

[0023] In a second aspect, an aluminum foil seal detection system based on deep learning is provided, the system comprising:

[0024] Image data acquisition module: used to acquire a large amount of infrared image data of the heated aluminum foil seal collected by the infrared thermal imager, and to acquire the sealing condition of the aluminum foil seal;

[0025] Image data processing module: used for preprocessing the infrared image data of the aluminum foil seal, and marking the preprocessed infrared image data of the aluminum foil seal with corresponding sealing conditions;

[0026] Detection network model training module: used to input the labeled aluminum foil seal infrared image data into the aluminum foil seal detection network model for parameter training and testing;

[0027] Aluminum foil seal detection module to be tested: used to obtain the infrared image of the aluminum foil seal to be tested, and input the infrared image of the aluminum foil seal to be tested into the trained aluminum foil seal detection network model to obtain the sealing condition.

[0028] In a third aspect, a computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the aluminum foil seal detection method based on deep learning described in the first aspect are implemented.

[0029] In a fourth aspect, a computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of an aluminum foil seal detection method based on deep learning as described in the first aspect above.

[0030] This application has at least the following beneficial effects:

[0031] The present application obtains a large number of infrared images of aluminum foil seals, opens the bottle cap to confirm the aluminum foil seal, pre-processes the infrared images of the aluminum foil seals, and uses the confirmed aluminum foil seals to mark them, uses the marked infrared images of the aluminum foil seals to train and test the detection network model, inputs the infrared images of the aluminum foil seals to be tested into the trained detection network model, and outputs the sealing level. Therefore, through the present application, the sealing status of the aluminum foil seal can be quickly and accurately obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1A schematic diagram of a flow chart of an aluminum foil seal detection method based on deep learning provided in one embodiment of the present application;

[0033] Figure 2 A detailed flowchart of an aluminum foil seal detection method based on deep learning provided in one embodiment of the present application;

[0034] Figure 3 A diagram showing the structure of a network model of an aluminum foil seal detection network provided in one embodiment of the present application;

[0035] Figure 4 A schematic diagram of the ConvBlock and CSPBlock_X structures provided for an embodiment of the present application;

[0036] Figure 5 A schematic diagram of the reasoning of the aluminum foil seal detection network model provided for one embodiment of the present application;

[0037] Figure 6 A lossless conversion effect diagram provided by an embodiment of the present application;

[0038] Figure 7 An image data enhancement effect diagram is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0040] In one embodiment, Figure 1 and Figure 2 As shown, a method for detecting aluminum foil seals based on deep learning is provided, comprising the following steps:

[0041] Step S1: obtaining a large amount of infrared image data of the heated aluminum foil seal collected by an infrared thermal imager, and obtaining the sealing condition of the aluminum foil seal.

[0042] In step S1, an infrared thermal imager is used to collect 16-bit original infrared image data after being heated by an aluminum foil sealing machine. The bottle cap is opened to confirm the aluminum foil sealing condition, and the data is matched one by one with the collected image data to construct an original aluminum foil infrared image database.

[0043] Step S2: preprocessing the aluminum foil sealing infrared image data, and marking the preprocessed aluminum foil sealing infrared image data with corresponding sealing conditions.

[0044] In step S2, the collected surgical instrument data is cleaned, denoised, losslessly converted, labeled, enhanced, and divided into data sets. The specific steps are as follows:

[0045] (1) Denoising data means using Gaussian filtering to remove noise points in an image. The specific operation is to use a Gaussian kernel to perform a convolution operation with each point in the input image, and then sum the convolution results to obtain the output image.

[0046] (2) Lossless conversion means that the 16-bit infrared image data is normalized and converted into an 8-bit RGB3 channel color image according to the color mapping table. Figure 6 shown.

[0047] (3) Annotate the preprocessed image data. The annotated data is the minimum circumscribed rectangle of the aluminum foil imaging in the image and the category of the sealing. The sealing category is determined by label_i in S1.

[0048] (4) In data enhancement, image data is flipped, rotated, randomly cropped, mixed upsampled, and CutMix data enhancement techniques are used to increase data diversity and improve the generalization ability of the model. Image enhancement effects are as follows: Figure 7 shown.

[0049] (5) Dataset division: The data processed in the above steps are randomly divided into training set, validation set and test set. The three data sets do not overlap with each other to ensure the authenticity of the algorithm test results. The data set implemented this time contains 9 categories, namely OK products and 8 categories of NG products such as virtual seals, folded edges, and reversed bottle caps. There are 2,000 samples of OK products and 1,000 samples of each category of NG products, for a total of 9,000 samples. The data set is divided into 8:1:1, and a total of 7,200 samples are randomly selected as training sets; 900 samples are randomly selected as validation sets; and the remaining 900 samples are used as test sets. The training set, validation set and test set do not overlap with each other.

[0050] At present, most commonly used aluminum foil seal detection methods based on computer vision use the color image given by the infrared camera to convert into a grayscale image for judgment. Since the internal working principle of the infrared camera is to convert the original depth data into a color image, the algorithm then converts the color image given by the infrared camera into a grayscale image, which is equivalent to two conversions, and the direct grayscale conversion of the color image will lose most of the data information. The preprocessing method based on lossless conversion proposed in this application only performs one data conversion, which saves preprocessing time compared to two conversions, and lossless conversion can enhance the contrast and expression ability of the data while ensuring that the data information is not lost, making the subsequent algorithm more accurate and reliable in the performance of aluminum foil seal detection.

[0051] Step S3: inputting the labeled aluminum foil seal infrared image data into the aluminum foil seal detection network model for parameter training and testing;

[0052] In step S3, the aluminum foil seal detection model includes an input layer, a backbone network (Backbone) improved based on CSPDarknet53, a path aggregation network (Path Aggregation Network, PANet), a detection head (DetectionHead) and a classification optimization module.

[0053] 1. The input layer receives and preprocesses the input image. The input image size is 640×640×3 and is normalized.

[0054] 2. The improved backbone network adopts the CSPDarknet53 structure. The overall network structure is as follows Figure 3 As shown in the figure, it combines the characteristics of EfficientNet to improve the feature extraction capability. It mainly includes six feature extraction modules (Module_1, ..., Module_6) and SPP modules. Each feature extraction module contains multiple ConvBlocks and CSPBlock_X. CSPBlock_X is composed of 3 ConvBlocks and X Resunit modules. The structure of ConvBlock and CSPBlock_X is as follows Figure 4 As shown, the details are as follows:

[0055] Specifically, the input image size of Module_1 is 640×640×3, which contains a ConvBlock, which is mainly composed of Conv2D, Batch Normalization and SiLU activation functions. The convolution kernel size is 6x6, the step size is 2, and the output feature map size is 320×320×32. Compared with the traditional CSPDarknet53, Module1 uses a 6x6 convolution kernel instead of the usual 3x3 convolution kernel to capture a wider range of feature information. The step size is 2, which halves the feature map size, reduces the amount of calculation of subsequent layers, and improves the computational efficiency of the entire network. Use SiLU (Sigmoid-Weighted LinearUnit) activation function instead of traditional ReLU. SiLU can effectively alleviate the gradient vanishing problem and enhance the nonlinear expression ability of the model through its smooth activation function curve.

[0056] Module_1, as the first layer of the entire network, is responsible for extracting preliminary feature information from the original input image. These features include basic features such as edges, textures, and color changes, which provide a basis for subsequent deeper feature extraction. Through a convolution operation with a step size of 2, Module_1 reduces the size of the input image from 640x640x3 to 320x320x32. Such a size reduction not only reduces the computational complexity of subsequent layers, but also enhances the network's ability to focus on local features. The use of the SiLU activation function enables Module_1 to better handle nonlinear changes in input features. This is especially important for target detection and classification in complex backgrounds.

[0057] Module2 and Module3 are CSPBlock_1 and CSPBlock_2 respectively, which are responsible for extracting more mid-level features and reducing the size of the feature map from 320×320×32 to 80*80*128.

[0058] Module4 and Module5 are both composed of CSPBlock_8, which is responsible for deeper feature extraction of data and extracts deeper features through a large number of residual connections. The feature map sizes after Module4 and Module5 are 40×40×256 and 20×20×512 respectively. Module6 is composed of CSPBlock_4, and the output feature map size is 10×10×1024. It extracts and refines the features extracted by the previous modules, and combines the outputs of 3 ConvBlock and 4 Resunit modules for feature fusion to ensure accurate recognition of the target.

[0059] Following Module6 is SPP (Spatial Pyramid Pooling), a module containing maximum pooling layers of different sizes (2, 3, 5), which is used to perform spatial pyramid pooling on the feature maps extracted by the previous residual module to enhance the feature extraction capability.

[0060] 3. The Path Aggregation Network (PANet) achieves effective fusion of multi-scale features through top-down and bottom-up paths, enhancing the model's ability to detect objects at different sizes and scales. Its connection with CSPDarknet53 is as follows:

[0061] (1) CSPDarknet53 outputs three feature maps of different scales, which come from different CSPBlock modules (Module4, Module5, Module6).

[0062] (2) After the feature map is input into PANet:

[0063] a) The first feature map is input to the first Concat+Conv x5 module of PANet.

[0064] b) The second feature map is input to the second Concat+Conv x5 module of PANet.

[0065] c) The third feature map is input to the third Concat+Conv x5 module of PANet.

[0066] (3) Top-Down Pathway:

[0067] a) Starting from the highest-level feature map (from the last Resblock_body of CSPDarknet53), upsample (Upsampling) and merge (Concat) with the feature map of the second highest level.

[0068] b) Continue upsampling and fusing with the next shallower feature map until all feature maps are processed.

[0069] (4) Bottom-Up Pathway:

[0070] a) Starting from the lowest layer feature map (from the first Resblock_body of CSPDarknet53), downsample (Downsampling) after convolution and fused (Concat) with the feature map of the second lowest layer.

[0071] b) Continue downsampling and fusing with the next deeper feature map until all feature maps are processed.

[0072] Through the top-down path, the deep feature map is gradually upsampled and fused with the shallow feature map, so that the high-level semantic information is passed to the lower layer. Through the bottom-up path, the shallow feature map is gradually downsampled and fused with the deep feature map, so that the low-level detail information is passed to the higher layer. This path aggregation mechanism enables the model to more effectively integrate multi-scale feature information and improve the detection accuracy and robustness of aluminum foil heat sealing under different ambient temperatures and backgrounds.

[0073] In the detection network of the present invention, the detection head performs target detection at different scales and outputs the position and category of the aluminum foil. Specifically, the detection head includes three output layers for detecting targets of small, medium and large scales. The specific structure of each output layer is as follows:

[0074] (1) Convolutional layer (ConvLayer): responsible for feature extraction and dimensionality reduction.

[0075] (2) Batch Normalization Layer: used to stabilize the training process, accelerate convergence and prevent overfitting.

[0076] (3) Swish Activation Function Layer: Enhances nonlinearity and improves the expressiveness of the model.

[0077] 4. Each output layer of the detection head uses the Swish activation function. Compared with the Leaky ReLU in YOLOv5, Swish can provide better nonlinear expression capabilities, so that the detection head of the present invention shows higher accuracy and robustness in target detection tasks, especially the detection effect of the aluminum foil sealing position and category is significantly improved.

[0078] 5. The model classification optimization module improves the classification accuracy and model robustness through the following technical means:

[0079] (1) Loss function optimization: A combination of Focal Loss and Cross-Entropy Loss is used to calculate the error between the predicted result and the true label, aiming to improve the classification accuracy, especially for small objects and objects that are difficult to classify.

[0080] (2) Data enhancement: Through data enhancement techniques (such as hybrid upsampling and CutMix), the diversity of data is increased and the generalization ability of the model is improved.

[0081] (3) Regularization: A Dropout layer is inserted between the fully connected layer and the convolutional layer of the model to prevent overfitting. Batch Normalization is applied after each convolutional layer and fully connected layer to improve the training speed and stability of the model.

[0082] The above-mentioned aluminum foil seal detection classification network model is trained by inputting the training data set obtained in data preprocessing into the neural network for multiple rounds of training. In each round of training, the model parameters are updated according to the aforementioned optimization algorithm and loss function. After multiple iterations and optimizations, a network model with the best performance on the validation set is finally obtained.

[0083] Next, we conduct data testing. Data testing uses the optimal network model obtained in the network training phase to infer the new data to determine the specific location and category of the aluminum foil. First, load the optimal network model obtained by S4 training into the inference environment. Then preprocess the input test data. This test uses the test set divided in S2, totaling 900 images. Finally, use the network model obtained by S4 training to perform model inference to obtain the location and category of the aluminum foil. The network inference diagram is as follows: Figure 5The classification standard during the test is based on OK products and NG products. OK products have good sealing, and NG products have poor sealing. The following table 1 shows the data test results.

[0084] Table 1

[0085] category Number of truth values The algorithm identifies the correct number Accuracy (%) OK products 200 198 99 NG products 700 694 99.14

[0086] Step S4: obtaining an infrared image of the aluminum foil seal to be tested, and inputting the infrared image of the aluminum foil seal to be tested into a trained aluminum foil seal detection network model to obtain a sealing level.

[0087] The aluminum foil seal detection classification network model constructed in this application has the following points:

[0088] (1) Most of the current aluminum foil seal inspection methods based on infrared imaging rely on traditional image processing technologies, such as edge detection, threshold segmentation, template matching, etc. These methods cannot meet the requirements of aluminum foil seal inspection when faced with complex temperature distribution and diverse defect forms. The solution provided by this application is to use a convolutional neural network to directly extract features from the non-destructive thermal imaging data of the aluminum foil seal, and combine it with a multi-scale feature fusion mechanism, using PANet (path aggregation network) to achieve effective fusion of multi-scale features through top-down and bottom-up paths. The top-down path performs feature fusion through upsampling, so that high-level semantic information is transmitted to the lower layer; the bottom-up path performs feature fusion through downsampling, so that low-level detail information is transmitted to the upper layer. Through this path aggregation mechanism, the present invention can more effectively integrate multi-scale feature information and improve the detection accuracy and robustness of aluminum foil seal defects under different sizes and backgrounds. Compared with traditional image processing technology, the present invention utilizes the powerful feature extraction capability of convolutional neural networks, can better cope with complex temperature distribution and diverse defect forms, and makes seal inspection more accurate and reliable.

[0089] (2) This application introduces regularization techniques, such as Dropout and Batch Normalization, to prevent model overfitting and improve the robustness of the model. In addition, data enhancement techniques, such as hybrid upsampling and CutMix, are used to increase the diversity of training data, further improving the generalization ability of the model and making it more stable and reliable in practical applications.

[0090] (3) This application optimizes the loss function by using a combination of Focal Loss and Cross-Entropy Loss, especially for small targets and targets that are difficult to classify. Focal Loss effectively reduces the impact of class imbalance on model training by adjusting the loss weights of samples of different categories, improves the detection accuracy of small defects and targets that are difficult to classify, and significantly improves the overall classification performance of the model.

[0091] (4) In the design of the detection head for sealing defects, the present invention realizes the detection and classification of features of different scales through multi-layer convolution, batch normalization and activation function (Swish). Compared with the traditional detection head design, the detection head of the present invention performs target detection at different scales, outputs the position and category of aluminum foil, can locate and classify sealing defects more accurately, and improves the efficiency and accuracy of the detection system.

[0092] In the above-mentioned aluminum foil seal detection method based on deep learning, a large number of infrared images of aluminum foil seals are obtained, and the bottle caps are opened to confirm the aluminum foil sealing conditions, the infrared images of the aluminum foil seals are preprocessed, and the confirmed aluminum foil sealing conditions are used for annotation, and the detection network model is trained and tested using the annotated infrared images of the aluminum foil seals, and the infrared images of the aluminum foil seals to be tested are input into the trained detection network model, thereby outputting the sealing level. Therefore, through this application, the sealing conditions of the aluminum foil seals can be quickly and accurately obtained.

[0093] In one embodiment, a deep learning-based aluminum foil seal detection system is provided, the system comprising:

[0094] Image data acquisition module: used to acquire a large amount of infrared image data of the heated aluminum foil seal collected by the infrared thermal imager, and to acquire the sealing condition of the aluminum foil seal;

[0095] Image data processing module: used for preprocessing the infrared image data of the aluminum foil seal, and marking the preprocessed infrared image data of the aluminum foil seal with corresponding sealing conditions;

[0096] Detection network model training module: used to input the labeled aluminum foil seal infrared image data into the aluminum foil seal detection network model for parameter training and testing;

[0097] Aluminum foil seal detection module to be tested: used to obtain the infrared image of the aluminum foil seal to be tested, and input the infrared image of the aluminum foil seal to be tested into the trained aluminum foil seal detection network model to obtain the sealing condition.

[0098] For the specific definition of an aluminum foil seal detection system based on deep learning, please refer to the definition of an aluminum foil seal detection method based on deep learning above, which will not be repeated here. Each module in the above-mentioned aluminum foil seal detection system based on deep learning can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0099] In one embodiment, a computer device is provided, which may be a server. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above-mentioned aluminum foil seal detection method based on deep learning is implemented.

[0100] In one embodiment, a computer program product is also provided, including a computer program / instruction, which, when executed by a processor, involves all or part of the process in the above-mentioned embodiment method.

[0101] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0102] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0103] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A deep learning-based aluminum foil seal detection method, characterized in that: The method comprises: Acquire a large amount of infrared image data of the heated aluminum foil seal collected by an infrared thermal imager, and obtain the sealing condition of the aluminum foil seal; Preprocessing the aluminum foil sealing infrared image data, and marking the preprocessed aluminum foil sealing infrared image data with corresponding sealing conditions; The labeled aluminum foil seal infrared image data is input into the aluminum foil seal detection network model for parameter training and testing; An infrared image of the aluminum foil seal to be tested is obtained, and the infrared image of the aluminum foil seal to be tested is input into a trained aluminum foil seal detection network model to obtain a sealing condition.

2. The aluminum foil seal detection method based on deep learning according to claim 1, characterized in that: The preprocessing includes: cleaning, denoising, lossless conversion, and data enhancement.

3. The method according to claim 1, characterized in that The sealing detection network model includes: an input layer, a backbone network, a path aggregation network, a detection head and a classification optimization module.

4. The aluminum foil seal detection method based on deep learning according to claim 3 is characterized in that: The backbone network adopts the CSPDarknet53 structure, including six feature extraction modules and one SPP module. The feature extraction module is composed of multiple ConvBlocks and CSPBlock_X; The feature extraction module is used to extract primary features, middle-level features, and deep-level features, and perform feature fusion on the primary features, middle-level features, and deep-level features; The SPP module includes maximum pooling layer modules of different sizes, and performs spatial pyramid pooling on the feature map extracted by the residual module in the feature extraction module.

5. The aluminum foil seal detection method based on deep learning according to claim 3, characterized in that: The path aggregation network includes upsampling, downsampling, convolution, feature fusion and skip connection, and fuses multi-scale features through top-down and bottom-up paths.

6. The aluminum foil seal detection method based on deep learning according to claim 3, characterized in that: The detection head includes three output layers for object detection at different scales and outputting the aluminum foil sealing level; The output layer includes a convolutional layer, a batch normalization layer, and an activation function.

7. The aluminum foil seal detection method based on deep learning according to claim 3, characterized in that: The classification optimization module uses loss function optimization, data enhancement and regularization; the loss function optimization adopts Focal Loss and Cross-Entropy Loss; the regularization uses Dropout and BatchNormalization.

8. An aluminum foil seal detection system based on deep learning, characterized in that: The system comprises: Image data acquisition module: used to acquire a large amount of infrared image data of the heated aluminum foil seal collected by the infrared thermal imager, and to acquire the sealing condition of the aluminum foil seal; Image data processing module: used for preprocessing the infrared image data of the aluminum foil seal, and marking the preprocessed infrared image data of the aluminum foil seal with corresponding sealing conditions; Detection network model training module: used to input the labeled aluminum foil seal infrared image data into the aluminum foil seal detection network model for parameter training and testing; Aluminum foil seal detection module to be tested: used to obtain the infrared image of the aluminum foil seal to be tested, and input the infrared image of the aluminum foil seal to be tested into the trained aluminum foil seal detection network model to obtain the sealing condition.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images