Ceramic wine bottle inner wall defect detection method and equipment

Abstract

The invention provides a ceramic wine bottle inner wall defect detection method and equipment, which is applied to the field of automatic detection, and comprises the following steps: collecting each inner wall image of a to-be-detected ceramic wine bottle; preprocessing each inner wall image to obtain a processed image corresponding to each inner wall image; inputting each processed image into a pre-trained detection model, enabling the detection model to perform inner wall defect analysis on each processed image, and outputting an inner wall defect detection result of the ceramic wine bottle; the detection model takes EPSANet as a backbone network and introduces a model of a feature pyramid network. By applying the scheme provided by the invention, after the inner wall image of the ceramic wine bottle is acquired, the trained detection model is used for performing inner wall defect analysis on the inner wall image, so that the inner wall defect detection result of the ceramic wine bottle is obtained, the defect of the inner wall of the wine bottle does not need to be detected manually, the condition of missing detection is avoided, and the detection accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of automated detection technology, and in particular to a method and equipment for detecting defects on the inner wall of a ceramic wine bottle. Background Art

[0002] In the traditional production and inspection process of ceramic wine bottles, quality inspection mainly relies on manual visual inspection. This method has many limitations, including but not limited to slow inspection speed, high labor intensity, strong subjectivity and high missed inspection rate. Especially when inspecting the inner wall of the bottle, due to the narrow bottle mouth, manual inspection is difficult to fully cover, resulting in limited inspection efficiency and accuracy. In addition, manual inspection requires high vision and attention of workers, and long working hours can easily lead to fatigue, further affecting the inspection quality. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a method and device for detecting defects on the inner wall of a ceramic wine bottle. By applying the solution provided in the present application, an image of the inner wall of the ceramic wine bottle is collected, the inner wall image is processed to obtain a processed image, and a trained detection model is used to analyze the inner wall defects of the processed image to obtain a detection result of the inner wall defects of the ceramic wine bottle. There is no need for manual defect detection, which avoids missed detections, improves the accuracy of the detection results, improves detection efficiency, and reduces detection costs.

[0004] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0005] A method for detecting defects on the inner wall of a ceramic wine bottle, comprising:

[0006] Collect images of each inner wall of the ceramic wine bottle to be inspected;

[0007] Preprocessing each of the inner wall images to obtain a processed image corresponding to each of the inner wall images;

[0008] Each of the processed images is input into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model uses EPSANet as the backbone network and introduces a feature pyramid network.

[0009] In the above method, optionally, the step of acquiring the inner wall image of the ceramic wine bottle to be inspected includes:

[0010] The lighting conditions of the ceramic wine bottle are controlled, and the inner wall images of the ceramic wine bottle at different angles under various lighting conditions are collected.

[0011] In the above method, optionally, the preprocessing of each inner wall image to obtain a processed image corresponding to each inner wall image includes:

[0012] Grayscale processing, filtering and denoising processing, and contrast enhancement processing are performed on each of the inner wall images to obtain a processed image corresponding to each of the inner wall images.

[0013] In the above method, optionally, the step of causing the detection model to perform inner wall defect analysis on each of the processed images and outputting the inner wall defect detection result of the ceramic wine bottle includes:

[0014] The detection model extracts inner wall feature data of each processed image, analyzes each inner wall feature data, and obtains analysis information of each inner wall image. When the analysis information includes an identification of a defect identified from the inner wall image, the analysis information includes description information of each defect identified in the inner wall image, and the description information includes a location, type, and confidence level of the defect.

[0015] The various analysis information are aggregated into inner wall defect detection results and output.

[0016] The above method may optionally further include:

[0017] Analyze the inner wall defect detection results to obtain various description information;

[0018] For each of the description information, the confidence in the description information is compared with the preset confidence. When the confidence in the description information is greater than the preset confidence, it is determined that the ceramic wine bottle has a defect corresponding to the description information.

[0019] The above method, optionally, includes the following steps of training the detection model:

[0020] Collecting a training data set and a verification data set; the training data set includes each inner wall sample image that has been processed with data enhancement;

[0021] Inputting each inner wall sample image in the training data set into a preset neural network model, so that the neural network model outputs a defect recognition result of each inner wall sample image;

[0022] Determining a defect label of each of the inner wall sample images, and determining function loss information of a loss function of the neural network model based on the defect label of each of the inner wall sample images and a defect recognition result;

[0023] Using the function loss information to update the model parameters of the neural network model, and then returning to the step of inputting each inner wall sample image in the training data set into the neural network model so that the neural network model outputs a defect recognition result of each inner wall sample image, until a preset stop condition is met, the training of the neural network model is stopped to obtain a trained neural network model;

[0024] The performance of the trained neural network model is evaluated using the validation data set, and the network layer parameters and / or network structure of the trained neural network model are optimized based on the evaluation results to obtain an optimized neural network model, and the optimized neural network model is determined as a detection model.

[0025] A ceramic wine bottle inner wall defect detection device, comprising:

[0026] Loading and unloading modules, material in and out modules, image acquisition modules and defect detection modules;

[0027] The loading and unloading modules are used to place ceramic wine bottles;

[0028] The material inlet and outlet module is used for transplanting the ceramic wine bottles;

[0029] The image acquisition module is used to acquire images of each inner wall of the ceramic wine bottle;

[0030] The defect detection module is used to preprocess each of the inner wall images to obtain a processed image corresponding to each of the inner wall images; each of the processed images is input into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model is a model that uses EPSANet as the backbone network and introduces a feature pyramid network.

[0031] In the above device, optionally, the image acquisition module comprises:

[0032] A probe that integrates multiple high-precision cameras and lighting units;

[0033] The high-precision camera is arranged at the lower end and the side of the probe rod;

[0034] The probe rod is connected to a rotating motor and a synchronous belt gear, so that the probe rod can rotate 360 ​​degrees to collect all-round information about the inner wall of the ceramic wine bottle.

[0035] In the above-mentioned device, optionally, the defect detection module executes a process in which the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle, including:

[0036] The detection model extracts inner wall feature data of each processed image, analyzes each inner wall feature data, and obtains analysis information of each inner wall image. When the analysis information includes an identification of a defect identified from the inner wall image, the analysis information includes description information of each defect identified in the inner wall image, and the description information includes a location, type, and confidence level of the defect.

[0037] The various analysis information are aggregated into inner wall defect detection results and output.

[0038] The above-mentioned device may optionally further include: an analysis module;

[0039] The analysis module is used to parse the inner wall defect detection results and obtain various descriptive information; for each descriptive information, the confidence in the descriptive information is compared with a preset confidence, and when the confidence in the descriptive information is greater than the preset confidence, it is determined that the ceramic wine bottle has a defect corresponding to the descriptive information.

[0040] Compared with the prior art, this application has the following advantages:

[0041] The present application provides a method and device for detecting inner wall defects of ceramic wine bottles, including: collecting each inner wall image of the ceramic wine bottle to be detected; preprocessing each inner wall image to obtain a processed image corresponding to each inner wall image; inputting each processed image into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each processed image, and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model is a model that uses EPSANet as a backbone network and introduces a feature pyramid network. Applying the solution provided by the present application, after collecting the inner wall image of the ceramic wine bottle, the inner wall defect analysis is performed on the inner wall image using the trained detection model, so as to obtain the inner wall defect detection result of the ceramic wine bottle, thereby eliminating the need for manual detection of the defects on the inner wall of the wine bottle, avoiding missed detection, and improving the accuracy and efficiency of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0043] Figure 1 A flow chart of a method for detecting defects on the inner wall of a ceramic wine bottle provided in an embodiment of the present application;

[0044] Figure 2A flow chart of a method for training a model provided in an embodiment of the present application;

[0045] Figure 3 Another flow chart of a method for detecting defects on the inner wall of a ceramic wine bottle provided in an embodiment of the present application;

[0046] Figure 4 A schematic structural diagram of a ceramic wine bottle inner wall defect detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0049] Terminology explanation:

[0050] EPSANet: is an open source deep learning project based on Python and PyTorch framework, which aims to provide a novel attention mechanism to improve the performance of the model in image classification and object detection tasks while maintaining low computational complexity and parameter count.

[0051] CenterNet model: is an Anchor-Free based target detection model, which has significant advantages in speed and accuracy.

[0052] As can be seen from the background technology, ceramic wine bottles need to be inspected when they are produced. The current inspection method relies on manual inspection, which has low inspection efficiency and is prone to errors.

[0053] In order to solve the above problems, the present application provides a ceramic wine bottle inner wall defect detection solution. The solution provided by the present application is applied to collect the inner wall image of the ceramic wine bottle, process the inner wall image, and then use EPSANet as the backbone network and introduce the detection model of the feature pyramid network to analyze the inner wall defects of the processed inner wall image, and then output the inner wall defect detection result of the ceramic wine bottle. The whole process does not require staff to perform inspections, avoiding the subjectivity of manual inspections and avoiding missed inspections during manual inspections. Using the trained detection model for inspections can improve the inspection efficiency and the accuracy of the inspection results.

[0054] The present application can be used in many general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, distributed computing environments including any of the above devices or equipment, etc.

[0055] Reference Figure 1 , which is a flow chart of a ceramic wine bottle inner wall defect detection method provided in an embodiment of the present application, as shown in S101-S103, and the specific description is as follows:

[0056] S101, collecting images of the inner walls of the ceramic wine bottle to be inspected.

[0057] Use a high-precision camera to capture images of the inner walls of the ceramic wine bottles to be inspected.

[0058] When collecting images of the inner walls of the ceramic wine bottle, the lighting conditions of the ceramic wine bottle are controlled, and pictures of the inner walls of the ceramic wine bottle at various angles under various lighting conditions are collected.

[0059] Lighting conditions can be understood as lighting brightness.

[0060] Collect pictures of the inner wall of the ceramic wine bottle at various angles under different lighting brightness. Preferably, the high-precision camera can capture 360-degree images of the inner wall of the ceramic wine bottle during the collection, and the images of the inner wall of the ceramic wine bottle include images of different parts such as the bottleneck, body and bottom of the ceramic wine bottle.

[0061] S102: Preprocess each inner wall image to obtain a processed image corresponding to each inner wall image.

[0062] Each inner wall image is subjected to grayscale processing, filtering and denoising processing, and contrast enhancement processing to obtain a processed image corresponding to each inner wall image.

[0063] Grayscale processing of the inner wall image can reduce the number of pixels in the image, simplify the information in the image, increase the processing speed, reduce the storage requirement, and highlight the features in the image. The grayscale processing methods in this application include: component method, maximum value method, average value method, weighted average method, YUV color space conversion method, etc.

[0064] Filtering and denoising the grayscaled image can remove high-frequency noise in the image, make the image clearer, help improve the visual quality of the image, and enhance the contrast between the target and the background in the image, which helps to identify the target content. It can also smooth the details in the image, reduce the texture and fine structure of the image, and retain edge information, which helps to maintain the structure and details of the image while enhancing the image, making the image more realistic. The filtering and denoising methods in this application include filter-based denoising methods, but are not limited to mean filtering, median filtering, Gaussian filtering, bilateral filtering, and non-local mean filtering; preferably, this application can also use model-based denoising methods, including but not limited to methods based on low-rank matrix approximation and methods based on sparse representation; this application can also use learning-based denoising methods including but not limited to convolutional neural network (CNN) denoising and deep residual network (ResNet) denoising.

[0065] Performing contrast enhancement on the filtered and denoised image can increase the difference between different brightness levels in the image, making the image more distinct, with clear details, enhancing the image hierarchy, and helping to identify and analyze the content in the image. The contrast enhancement methods in this application include histogram stretching, histogram equalization, and adaptive histogram equalization.

[0066] S103, input each processed image into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each processed image and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model uses EPSANet as the backbone network and introduces a feature pyramid network.

[0067] In the solution provided in the embodiment of the present application, the detection model extracts the inner wall feature data of each processed image, analyzes each inner wall feature data, and obtains analysis information of each inner wall image. When the analysis information includes the identification of defects recognized from the inner wall image, the analysis information includes description information of each defect recognized in the inner wall image, and the description information includes the location, type and confidence of the defect; the various analysis information are aggregated into an inner wall defect detection result and output.

[0068] Preferably, the inner wall characteristic data includes but is not limited to characteristic data such as transmittance, shadow, background, reflection, cracks, foreign matter, and bubbles.

[0069] The detection model adopts an improved CenterNet model. Specifically, EPSANet is used as the backbone network, and a pyramid network structure (FPN) is introduced. EPSANet is used as the backbone network to extract features, and then FPN is used to further process these features to construct a multi-scale feature pyramid, thereby enhancing the feature extraction ability of the model; the detection model takes into account the high reflective background when inspecting the inner wall of ceramic wine bottles, defects with high similarity to the background, and the compatibility of various ceramic wine bottle inner wall color defect detection.

[0070] In the process of analyzing the inner wall defects of each processed image, the detection model first extracts the inner wall feature data. The process of extracting the inner wall feature data is as follows: each processed image input to the detection model first passes through the benchmark network part of EPSANet, which usually includes a series of convolutional layers, activation functions, and possible downsampling operations. The PSA module in EPSANet plays a role in this process, extracting and integrating spatial information of different scales through the pyramid segmentation attention mechanism to enhance the expression ability of features. After being processed by EPSANet, a series of feature maps with rich semantic information will be obtained.

[0071] The feature map is then fed into the FPN structure, which fuses feature maps of different levels by combining top-down and bottom-up methods. The FPN in the detection model uses upsampling and lateral connections to combine high-level semantic information with low-level detail information to form a multi-scale feature pyramid. After this processing, the feature map of each scale contains different levels of semantic and spatial information, which is particularly suitable for detecting objects of different sizes.

[0072] Finally, the multi-scale feature maps are fed into the head network of target detection. In CenterNet, this usually involves predicting the center point coordinates of the target, the width and height of the target, and the category of the target. The Head part of CenterNet processes the feature pyramid generated by FPN and generates the final detection result, including bounding boxes, classification probabilities, etc. The final detection result here is the inner wall defect detection result mentioned above. Due to the addition of FPN, the model can more effectively detect targets of different sizes, especially performing better in small target detection.

[0073] It should be noted that the analysis information of the inner wall image under different lighting conditions is different. Preferably, when the analysis information of the inner wall image includes a mark indicating that no defects are identified from the inner wall image, it means that no defects are identified from the inner wall image. Preferably, there is no description information in the analysis information at this time. When the analysis information of the inner wall image includes a mark indicating that defects are identified from the inner wall image, it indicates that defects are identified from the inner wall image. At this time, the analysis information includes description information of the identified defects. Preferably, there are as many description information as there are identified defects. The description information includes the location, type and confidence of the identified defects. The location is the location of the identified defect on the ceramic wine bottle. The type is the type of defect, such as cracks, bubbles, foreign matter, stains, etc. The confidence is the credibility of the identified defects.

[0074] In the scheme provided by the present application, each inner wall image of the ceramic wine bottle to be detected is collected; each inner wall image is preprocessed to obtain a processed image corresponding to each inner wall image; each processed image is input into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each processed image and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model is a model that uses EPSANet as the backbone network and introduces a feature pyramid network. Applying the scheme provided by the present application, after collecting the inner wall image of the ceramic wine bottle, the inner wall defect analysis is performed on the inner wall image using the trained detection model, thereby obtaining the inner wall defect detection result of the ceramic wine bottle, thereby eliminating the need for manual detection of the defects on the inner wall of the wine bottle, avoiding missed detection, and improving the accuracy and efficiency of detection.

[0075] Preferably, after obtaining the inner wall defect detection results of the ceramic wine bottle, the inner wall defect detection results can also be analyzed to obtain various descriptive information; for each descriptive information, the confidence in the description information is compared with the preset confidence. When the confidence in the description information is greater than the preset confidence, it is determined that the ceramic wine bottle has a defect corresponding to the description information.

[0076] It should be noted that when the confidence in the description information is greater than the preset confidence, it is determined that there is a defect corresponding to the description information on the ceramic wine bottle. In other words, it can be determined that the defect corresponding to the description information is real, or it can be understood that the credibility of the identification result of the defect corresponding to the description information is very high, and it can be determined that the defect does exist on the ceramic wine bottle; when the confidence in the description information is not greater than the preset confidence, it is determined that the credibility of the identified defect corresponding to the description information is low, and the defect corresponding to the description information does not exist on the bottle. Furthermore, the preset confidence can be set according to actual needs and is not limited here.

[0077] By analyzing the inner wall defect detection results, the confidence level in the description information is used to further determine whether there are defects on the ceramic wine bottle, thereby obtaining more accurate results.

[0078] Reference Figure 2 , is a flow chart of a method for training a model provided in an embodiment of the present application, specifically as shown in S201-S207, and the specific description is as described:

[0079] S201, collecting a training data set and a verification data set; the training data set includes each inner wall sample image that has been processed with data enhancement.

[0080] It should be noted that the inner wall sample images in the training data set are preprocessed and labeled images. The labeled information includes but is not limited to the location, type and size of the defects. The labeled information is crucial for the training model to identify specific defects.

[0081] Preferably, the data enhancement processing methods of the present application include but are not limited to image rotation, scaling, flipping, etc. By performing data enhancement processing, the generalization ability of the model for defect recognition under different conditions can be enhanced.

[0082] Preferably, the verification data set includes multiple verification images. Preferably, the verification images may also be annotated images. The annotated contents are as described above and will not be repeated here.

[0083] S202: Input each inner wall sample image in the training data set into a preset neural network model, so that the neural network model outputs a defect recognition result of each inner wall sample image.

[0084] The neural network model adopts an improved CenterNet model, with EPSANet as the backbone network, and introduces a pyramid structure (FPN) to enhance feature extraction capabilities. The model also takes into account the high reflective background when inspecting the inner wall of ceramic wine bottles, defects with high similarity to the background, and the compatibility of various ceramic wine bottle inner wall color defect detection.

[0085] After the inner wall sample image is input into the neural network model, the neural network model outputs the defect recognition result of each inner wall sample image. The defect recognition result contains relevant information of the defects recognized on the inner wall sample image, such as the type, size, location and other contents of the defects.

[0086] S203, determining the defect labels of each inner wall sample image, and based on the defect labels of each inner wall sample image and the defect recognition results, determining the function loss information of the loss function of the neural network model.

[0087] Preferably, the defect label here is the information marked on the inner wall sample image described in S201.

[0088] Using the defect labels and defect recognition results of each inner wall sample image, function loss information of the loss function of the neural network model is determined, and the function loss information includes but is not limited to the loss value of the loss function or the gradient loss information.

[0089] S204: Use the function loss information to update the model parameters of the neural network model.

[0090] S205, determining whether the neural network model satisfies a preset stop condition. When the neural network model does not satisfy the stop condition, returning to execute S202; when the neural network model satisfies the stop condition, executing S206.

[0091] The preset stopping conditions include but are not limited to whether the number of iterations reaches the preset number, whether the neural network model converges, etc.

[0092] S206: Stop training the neural network model to obtain a trained neural network model.

[0093] S207. Use the verification data set to evaluate the performance of the trained neural network model, and optimize the network layer parameters and / or network structure of the trained neural network model based on the evaluation results to obtain an optimized neural network model, and determine the optimized neural network model as the detection model.

[0094] In the process of training the detection model, the present application first uses a training data set to train the neural network model, then uses a verification data set to evaluate the performance of the trained neural network model, and then optimizes the network layer parameters and / or network structure of the neural network model to improve the performance of the neural network model, so that the neural network model can achieve the expected effect, improve the detection accuracy of the neural network model, and finally determine the optimized neural network model as the detection model.

[0095] The present application embodiment also provides another flow chart of the method for detecting defects on the inner wall of a ceramic wine bottle. Figure 3 The relevant instructions are as follows:

[0096] S1, Model construction.

[0097] A neural network model for ceramic wine bottle inner wall defect detection is established. The neural network model adopts the improved CenterNet model, in which EPSANet is used as the backbone network and a pyramid structure (FPN) is introduced. The model also takes into account the high reflective background during ceramic bottle inner wall detection, defects with high similarity to the background, and the compatibility of various ceramic bottle inner wall color defect detection.

[0098] S2, training set construction.

[0099] Construct a training set for neural network training; use a high-precision camera to collect a certain number of images of the inner wall of ceramic wine bottles, including normal and defective samples. After preprocessing, these images are annotated to clarify the location, type, and size of defects, including foreign matter, cracks, stains, etc.; construct a rich and diverse training data set to ensure the effectiveness and robustness of model training.

[0100] The training set here is equivalent to the training data set above.

[0101] S3, model training and testing.

[0102] The neural network is trained based on the training set, and the parameters of the neural network are continuously adjusted and optimized to improve the generalization ability and detection accuracy of the model. Preferably, data enhancement technology is also used in the training process, such as image rotation, scaling, flipping, etc., to enhance the generalization ability of the model for defect recognition under different conditions. Preferably, data enhancement technology can be used when constructing the training set to obtain a variety of images.

[0103] After training is completed, the trained model is tested using the validation data set, and then the model is tuned, including adjusting the parameters of the network layer, optimizing the network structure, etc., to further improve the performance of the model; the content of the test of the trained model includes accuracy, recall rate and other directions to ensure that the model achieves the expected effect. It should be noted that testing the trained model is equivalent to the content of evaluating the performance of the model mentioned above. After testing the model, the model is tuned to ensure that the neural network model can maintain high accuracy and high recall in various complex environments to meet the needs of industrial-grade applications.

[0104] S4, device integration and model deployment.

[0105] An automated device suitable for inspecting the inner wall of ceramic wine bottles is deployed, and the trained neural network model is installed on the device. The device includes but is not limited to loading and unloading modules, material input and output modules, probe rod camera modules (equivalent to the image acquisition module obtained above) and defect detection modules.

[0106] The probe camera module of the present application includes a probe that can be inserted into the bottle mouth, and the probe integrates 5 high-precision micro cameras and a lighting unit. Considering the narrow space of the bottle mouth, the micro camera and the camera body are separated to adapt to the narrow bottle mouth environment. The device is also equipped with a small user interaction system, through which human-computer interaction can be carried out to achieve the purpose of automatically detecting defects on the inner wall of ceramic wine bottles.

[0107] S5, defect detection.

[0108] Automated equipment is used to detect defects on the inner wall of ceramic wine bottles. Specifically, the equipment automatically grabs the wine bottle through a robotic arm, inserts a probe rod with a high-precision micro camera and lighting unit into the mouth of the ceramic wine bottle for image sampling and preprocessing, and inputs the preprocessed image into the trained neural network model for detection. The detection results include the type and confidence of the defects. If the confidence is greater than a certain threshold, it is considered that there is a defect on the inner wall of the ceramic wine bottle.

[0109] Another example: First, a high-precision micro camera with multiple lenses and multiple angles is used to collect high-resolution images of the inner wall of a ceramic wine bottle. The acquisition process is carried out under controlled lighting conditions. After the image is preprocessed, including grayscale, filtering and denoising, contrast enhancement and other operations, it is input into the trained neural network model. The neural network model analyzes the image based on the learned features, identifies and locates the inner wall defects, and the detection results output by the model include the location, type and confidence of the defect. By setting a confidence threshold, the detection result is judged as a real defect only when the confidence exceeds the threshold.

[0110] This application combines automation equipment and deep learning algorithms, uses high-precision micro cameras and optimized neural network models, and realizes rapid detection of the inner wall of ceramic wine bottles. It can accurately identify and classify tiny defects on the inner wall of ceramic wine bottles, including defects that are difficult to detect with traditional manual detection. It overcomes the compatibility issues of high reflective background, defects with high background similarity, and various color defect detections on the inner wall of ceramic wine bottles during detection of the inner wall of ceramic wine bottles. The detection recognition rate of certain defects is close to 100%, which significantly reduces the time required for manual detection and improves the overall detection efficiency and accuracy. At the same time, it reduces the labor cost of ceramic wine bottle manufacturers and improves product quality. In the automated detection equipment and method provided by this application, by integrating advanced hardware and software technologies, efficient and accurate detection of defects on the inner wall of ceramic wine bottles is achieved, which has broad industrial application prospects.

[0111] The present application also provides a ceramic wine bottle inner wall defect detection device, which is used to support Figure 1 The specific implementation of the method shown. Figure 4 , is a schematic diagram of the structure of a ceramic wine bottle inner wall defect detection device provided in an embodiment of the present application, and is specifically described as follows:

[0112] The ceramic wine bottle inner wall defect detection device provided in the embodiment of the present application includes: a loading and unloading module 301, a material input and output module 302, an image acquisition module 303 and a defect detection module 304.

[0113] The loading and unloading module 301 is used for placing ceramic wine bottles.

[0114] The loading and unloading module includes two groups of loading and unloading stations. Preferably, each group is provided with 4 fixtures, and 4 ceramic wine bottles can be placed into the material inlet and outlet module each time for equipment detection.

[0115] The material inlet and outlet module is used for transplanting the ceramic wine bottles.

[0116] The material in and out module uses a precision pneumatic module to make the cylindrical belt and supporting guide rail work to achieve bottle transfer.

[0117] The image acquisition module is used to acquire images of the inner walls of the ceramic wine bottle.

[0118] It should be noted that the image acquisition model integrates a probe rod with multiple high-precision cameras and a lighting unit; high-precision cameras are set at the lower end and side of the probe rod; the probe rod is connected to a rotating motor and a synchronous belt gear, so that the probe rod can rotate 360 ​​degrees and collect all-round information about the inner wall of the ceramic wine bottle.

[0119] Exemplarily, the image acquisition module includes a probe rod optical module and a camera module. The probe rod has a diameter of 10 mm and integrates 5 groups of cameras and a whole low-glare light source (the inner cavity of the wine bottle is highly reflective and easily produces light spots). The side of the probe rod is equipped with 4 high-precision micro cameras and the bottom of the probe rod is equipped with 1 high-precision micro camera, which are used to shoot the side wall and bottom of the wine bottle respectively; and the probe rod is connected to a rotating motor and a synchronous belt gear to ensure that the probe rod can rotate 360 ​​degrees synchronously to form an image and capture the inner wall of the wine bottle in all directions.

[0120] The defect detection module is used to preprocess each inner wall image to obtain a processed image corresponding to each inner wall image; each processed image is input into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each processed image and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model is a model that uses EPSANet as the backbone network and introduces a feature pyramid network.

[0121] Furthermore, the defect detection module executes a process in which the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle. Specifically, the detection model extracts the inner wall feature data of each processed image, analyzes each inner wall feature data, and obtains analysis information of each inner wall image. When the analysis information includes an identification of a defect recognized from an inner wall image, the analysis information includes description information of each defect recognized in the inner wall image, and the description information includes the location, type and confidence of the defect; and each piece of analysis information is aggregated into an inner wall defect detection result and outputted.

[0122] Furthermore, the ceramic wine bottle inner wall defect detection equipment also includes: an analysis module; the analysis module is used to parse the inner wall defect detection results and obtain various descriptive information; for each descriptive information, the confidence in the description information is compared with the preset confidence, and when the confidence in the description information is greater than the preset confidence, it is determined that the ceramic wine bottle has a defect corresponding to the description information.

[0123] In the device provided in the embodiment of the present application, the defect detection module is composed of a model algorithm and a memory, a processor, a display, etc. If the inner wall of the ceramic wine bottle is detected to be free of defects, an OK light will be displayed, and if there are defects, an NG light will be displayed.

[0124] The operator can start the inspection process through the user interface of the ceramic wine bottle inner wall defect inspection equipment. The equipment automatically completes image acquisition, processing and defect detection, and displays the results in real time on the interface. The operator can further analyze and process the inspection results.

[0125] A ceramic wine bottle defect detection device provided in an embodiment of the present application realizes comprehensive, efficient and accurate automated detection of the inner wall of the ceramic wine bottle by integrating a high-precision camera module and an advanced deep learning algorithm, so as to achieve the purpose of improving production efficiency, reducing costs and improving product quality.

[0126] Although the present invention depicts operations in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.

[0127] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0128] An embodiment of the present invention further provides a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the above-mentioned ceramic wine bottle inner wall defect detection method.

[0129] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards in the relevant regions.

[0130] The specific implementation processes and derivative methods of the above-mentioned embodiments are all within the protection scope of the present invention.

[0131] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0132] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0133] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting defects on the inner wall of a ceramic wine bottle, characterized in that: include: Collect images of each inner wall of the ceramic wine bottle to be inspected; Preprocessing each of the inner wall images to obtain a processed image corresponding to each of the inner wall images; Each of the processed images is input into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model uses EPSANet as the backbone network and introduces a feature pyramid network.

2. The method according to claim 1, characterized in that The step of collecting the inner wall image of the ceramic wine bottle to be inspected includes: The lighting conditions of the ceramic wine bottle are controlled, and the inner wall images of the ceramic wine bottle at different angles under various lighting conditions are collected.

3. The method according to claim 1, characterized in that: The preprocessing of each of the inner wall images to obtain a processed image corresponding to each of the inner wall images includes: Grayscale processing, filtering and denoising processing, and contrast enhancement processing are performed on each of the inner wall images to obtain a processed image corresponding to each of the inner wall images.

4. The method according to claim 1, characterized in that The step of making the detection model perform inner wall defect analysis on each of the processed images and outputting the inner wall defect detection result of the ceramic wine bottle includes: The detection model extracts inner wall feature data of each processed image, analyzes each inner wall feature data, and obtains analysis information of each inner wall image. When the analysis information includes an identification of a defect identified from the inner wall image, the analysis information includes description information of each defect identified in the inner wall image, and the description information includes a location, type, and confidence level of the defect. The various analysis information are aggregated into inner wall defect detection results and output.

5. The method according to claim 4, characterized in that Also includes: Analyze the inner wall defect detection results to obtain various description information; For each of the description information, the confidence in the description information is compared with the preset confidence. When the confidence in the description information is greater than the preset confidence, it is determined that the ceramic wine bottle has a defect corresponding to the description information.

6. The method according to claim 1, characterized in that The process of training the detection model includes: Collecting a training data set and a verification data set; the training data set includes each inner wall sample image that has been processed with data enhancement; Inputting each inner wall sample image in the training data set into a preset neural network model, so that the neural network model outputs a defect recognition result of each inner wall sample image; Determining a defect label of each of the inner wall sample images, and determining function loss information of a loss function of the neural network model based on the defect label of each of the inner wall sample images and a defect recognition result; Using the function loss information to update the model parameters of the neural network model, and then returning to the step of inputting each inner wall sample image in the training data set into the neural network model so that the neural network model outputs a defect recognition result of each inner wall sample image, until a preset stop condition is met, the training of the neural network model is stopped to obtain a trained neural network model; The performance of the trained neural network model is evaluated using the validation data set, and the network layer parameters and / or network structure of the trained neural network model are optimized based on the evaluation results to obtain an optimized neural network model, and the optimized neural network model is determined as a detection model.

7. A ceramic wine bottle inner wall defect detection device, characterized in that: include: Loading and unloading modules, material in and out modules, image acquisition modules and defect detection modules; The loading and unloading modules are used to place ceramic wine bottles; The material inlet and outlet module is used for transplanting the ceramic wine bottles; The image acquisition module is used to acquire images of each inner wall of the ceramic wine bottle; The defect detection module is used to preprocess each of the inner wall images to obtain a processed image corresponding to each of the inner wall images; each of the processed images is input into a pre-trained detection model, so that the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle; the detection model is a model that uses EPSANet as the backbone network and introduces a feature pyramid network.

8. The device according to claim 7, characterized in that The image acquisition module comprises: A probe that integrates multiple high-precision cameras and lighting units; The high-precision camera is arranged at the lower end and the side of the probe rod; The probe rod is connected to a rotating motor and a synchronous belt gear, so that the probe rod can rotate 360 ​​degrees to collect all-round information about the inner wall of the ceramic wine bottle.

9. The device according to claim 7, characterized in that The defect detection module executes a process in which the detection model performs inner wall defect analysis on each of the processed images and outputs the inner wall defect detection result of the ceramic wine bottle, including: The detection model extracts inner wall feature data of each processed image, analyzes each inner wall feature data, and obtains analysis information of each inner wall image. When the analysis information includes an identification of a defect identified from the inner wall image, the analysis information includes description information of each defect identified in the inner wall image, and the description information includes a location, type, and confidence level of the defect. The various analysis information are aggregated into inner wall defect detection results and output.

10. The device according to claim 7, characterized in that Also includes: Analysis module; The analysis module is used to analyze the inner wall defect detection results and obtain various description information; For each of the description information, the confidence in the description information is compared with the preset confidence. When the confidence in the description information is greater than the preset confidence, it is determined that the ceramic wine bottle has a defect corresponding to the description information.

Images