Quartz weight online quality detection method and system based on improved YOLOv8

Abstract

The invention relates to the technical field of computer vision, and discloses an improved YOLOv8-based quartz weight online quality detection method and system, and the method comprises the steps: obtaining a production image of a quartz weight in a kiln, inputting the production image of the quartz weight into a hierarchical feature extraction backbone network, obtaining image features of different hierarchies, and carrying out the extraction of the image features of the different hierarchies; and inputting the image features of different levels into the feature enhancement and fusion network to obtain feature enhancement and fusion features, and inputting the feature enhancement and fusion features into the multi-scale feature pyramid network to obtain a multi-scale detection result. According to the invention, online quality detection can be carried out efficiently and accurately, various indexes in the quartz weight production process can be monitored in real time, the quality detection level of the quartz weight is improved, and the influence of human factors on the detection result is reduced.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and particularly to an online quality detection method and system for quartz blanks based on improved YOLOv8. Background Art

[0002] As an important raw material, quartz blanks are widely used in fields such as electronics, optics, and metallurgy. During the production process of quartz blanks, ensuring the stability and consistency of their surface quality is crucial. However, due to the high-temperature environment inside the kiln and the complexity of the production process, traditional manual detection methods mainly rely on manual observation through the observation window to observe the production status of quartz blanks. This method is not only inefficient but also easily affected by human factors, making it difficult to guarantee the accuracy of quality detection. Therefore, how to achieve real-time detection of quality problems during the production process of quartz blanks, especially surface defects (such as bubbles, nodules) and dimensions (such as blank surface height, hump ratio, etc.), has become an urgent problem to be solved in the industry.

[0003] With the rapid development of computer vision technology, object detection methods based on deep learning have achieved remarkable application results in the field of industrial detection. However, during the production process of quartz blanks, affected by factors such as the complex background inside the kiln and the small size of the target, the detection performance of existing models for small targets (such as bubbles, nodules) still has certain limitations. When using the traditional single OpenCV method for detection, although some features can be effectively extracted through simple image processing techniques (such as edge detection, contour analysis, threshold segmentation, etc.), its ability to identify small targets (such as bubbles, nodules) in complex scenarios is relatively limited. Especially in a complex production environment, the posture of the quartz blank and the change of the furnace temperature often affect the image quality, making the boundary of the target unclear, and even in some cases, the contrast between the target and the background is relatively low, resulting in the difficulty of traditional methods to achieve accurate detection.

[0004] In addition, simply relying on traditional image processing techniques cannot fully utilize the advantages of deep learning in feature extraction and pattern recognition. Especially when facing small targets or subtle defects, traditional methods usually have difficulty achieving high precision and cannot provide sufficient robustness, resulting in unsatisfactory detection effects.

[0005] Currently, YOLOv8 provides stronger accuracy and efficiency in object detection tasks, but its recognition of small targets and robustness in complex backgrounds still need to be improved. Therefore, there is an urgent need for a new model to improve the accuracy and real-time performance of quartz blank quality detection. Summary of the Invention

[0006] The purpose of the present invention is to overcome one or more of the above existing technical problems, and provide an online quality detection method and system for quartz blanks based on improved YOLOv8.

[0007] To achieve the above object, an on-line quality detection method for quartz ingots based on improved YOLOv8 provided by the present invention includes:

[0008] Obtain the production image of quartz ingots in the kiln;

[0009] Input the production image of quartz ingots into the hierarchical feature extraction backbone network to obtain image features at different levels;

[0010] Input the image features at different levels into the feature enhancement and fusion network to obtain feature enhancement and fusion features;

[0011] Input the feature enhancement and fusion features into the multi-scale feature pyramid network to obtain multi-scale detection results.

[0012] According to one aspect of the present invention, the hierarchical feature extraction backbone network includes five Ghost convolutions with a step size of 2 and a convolution kernel of 3×3, four cross-stage feature fusion modules, a pyramid pooling layer, and an attention mechanism layer;

[0013] Input the production image of quartz ingots in the kiln into the first Ghost convolution to obtain the first convolution feature, where the formula is,

[0014]

[0015] Among them, represents the Ghost convolution;

[0016] P1 represents the first convolution feature;

[0017] X represents the production image of quartz ingots in the kiln;

[0018] Input the first convolution feature into the second Ghost convolution and the first cross-stage feature fusion module in sequence to obtain the second convolution feature, where the formula is,

[0019]

[0020] Among them, C2f 3 represents the cross-stage feature fusion module with a repeated stacking times of 3;

[0021] P2 represents the second convolution feature;

[0022] Input the second convolution feature into the third Ghost convolution and the second cross-stage feature fusion module in sequence to obtain the third convolution feature, where the formula is,

[0023]

[0024] Among them, C2f 6It represents a cross-stage feature fusion module with a repeated stacking times of 6;

[0025] P3 represents the third convolutional feature;

[0026] The third convolutional feature is sequentially input into the fourth ghost convolution and the third cross-stage feature fusion module to obtain the fourth convolutional feature, where the formula is,

[0027]

[0028] where, P4 represents the fourth convolutional feature;

[0029] The fourth convolutional feature is sequentially input into the fifth ghost convolution, the fourth cross-stage feature fusion module, the pyramid pooling layer and the attention mechanism layer to obtain the fifth convolutional feature, where the formula is,

[0030]

[0031] where, P5 represents the fifth convolutional feature;

[0032] CBAM represents the attention mechanism layer;

[0033] SPPF represents the pyramid pooling layer;

[0034] C2f 3 It represents a cross-stage feature fusion module with a repeated stacking times of 3.

[0035] According to one aspect of the present invention, the feature enhancement and fusion network includes an upsampling fusion layer and a downsampling fusion layer;

[0036] The upsampling fusion layer includes three upsampling fusion modules, and each upsampling fusion module has the same architecture, which is composed of a bilinear upsampling operation layer, a fusion layer and a cross-stage feature fusion module;

[0037] The fifth convolutional feature and the fourth convolutional feature are input into the first upsampling fusion module to obtain the first fusion feature, where the formula is,

[0038]

[0039] where, Q1 represents the first fusion feature;

[0040] Concat represents the channel fusion mechanism;

[0041] It represents the bilinear upsampling operation layer;

[0042] The first fusion feature and the third convolutional feature are input into the second upsampling fusion module to obtain the second fusion feature, where the formula is,

[0043]

[0044] Among them, Q2 represents the second fusion feature;

[0045] Input the second fusion feature and the second convolutional feature into the third upsampling fusion module to obtain the third fusion feature, where the formula is

[0046]

[0047] Among them, Q3 represents the third fusion feature.

[0048] According to one aspect of the present invention, the downsampling fusion layer includes three downsampling fusion modules, and each downsampling fusion module has the same architecture, which is composed of a downsampling operation layer, a fusion layer, and a cross-stage feature fusion module;

[0049] Input the second fusion feature and the third fusion feature into the first downsampling fusion module to obtain the fourth fusion feature, where the formula is

[0050]

[0051] Among them, Q4 represents the fourth fusion feature;

[0052] represents the downsampling operation layer;

[0053] Input the fourth fusion feature and the first fusion feature into the second downsampling fusion module to obtain the fifth fusion feature, where the formula is

[0054]

[0055] Among them, Q5 represents the fifth fusion feature;

[0056] Input the fifth fusion feature and the fifth convolutional feature into the third downsampling fusion module to obtain the sixth fusion feature, where the formula is

[0057]

[0058] Among them, Q6 represents the sixth fusion feature.

[0059] According to one aspect of the present invention, the multi-scale feature pyramid network includes feature channel superposition, classification convolution, and regularization convolution;

[0060] Input the third fusion feature, the fourth fusion feature, the fifth fusion feature, and the sixth fusion feature into the multi-scale feature pyramid network respectively to obtain detection results of different scales, where the formula is

[0061]

[0062] Among them, M represents the multi-scale feature pyramid network;

[0063] Detect represents the detection results at different scales;

[0064] represents the classification convolution;

[0065] Q k represents different fused features;

[0066] represents the regular convolution;

[0067] represents the feature channel superposition;

[0068] H k represents the height of the feature map at scale k;

[0069] W k represents the width of the feature map at scale k;

[0070] c represents the object confidence;

[0071] (x, y, w, h) represents the bounding box coordinate information;

[0072] C represents the class probability distribution.

[0073] According to one aspect of the present invention, the bounding box coordinates of the surface of the bubble, nodule and quartz ingot are obtained based on the multi-scale detection results;

[0074] The detection area of each class is binarized according to the bounding box coordinates to remove background noise;

[0075] The Gaussian filter smoothing technique is used to denoise the image and the external contours of the surface of the bubble, nodule and quartz ingot are extracted based on the sub-pixel contour detection technique, and the height of the ingot surface, the hump ratio and the time, position and size of the bubble and nodule appearance are calculated.

[0076] To achieve the above object, the present invention provides a quartz ingot online quality detection system based on the improved YOLOv8, including:

[0077] A quartz ingot manufacturing furnace for producing quartz ingots;

[0078] An imaging detection device is arranged on the observation port of the quartz ingot manufacturing furnace to collect the online production data of the quartz ingot in the quartz ingot manufacturing furnace in real time;

[0079] A computer image processing device for receiving the production data of the imaging detection device and analyzing the production data;

[0080] The detection interface is connected to a computer image processing device and displays the detection results in real time;

[0081] The interface controller is connected to a computer image processing device and obtains the detection results in real time.

[0082] According to one aspect of the present invention, the interface controller includes a communication module, an alarm module, and a data recording module;

[0083] The communication module is used for data transmission with a computer image processing device;

[0084] The alarm module triggers an audible and visual alarm when an abnormal quality of the quartz ingot is detected, reminding the operator to handle it in time;

[0085] The data recording module records the detection data and time stamps and generates a data log file for later quality analysis and traceability.

[0086] Based on this, the beneficial effects of the present invention are as follows: By introducing the CBAM module and the ultra-small detection head, the present invention enhances the response value of key features through a dual attention mechanism and expands the detection scale to a resolution of 160×160, enhancing the recognition ability of small targets, significantly improving the detection accuracy in complex environments, especially in the case of more background noise, and being able to more accurately identify small targets such as bubbles and nodules;

[0087] The present invention proposes an improved method for detecting the quality of quartz ingots using YOLOv8. Through lightweight design by replacing the standard C3 module with the C2f module, the computational complexity is reduced, enabling the system to have the advantage of strong real-time performance while ensuring high precision. The present invention can monitor the production process of quartz ingots in real time, automatically identify defects and generate corresponding deviation curves, greatly improving the detection efficiency and automation level of the production line;

[0088] By introducing spatial and channel attention mechanisms, the present invention can effectively suppress background noise and improve the robustness of the model. Especially in complex backgrounds and low-contrast images, small targets can still be stably detected, thus overcoming the limitations of traditional methods in these environments. By obtaining the surface of the quartz ingot as the detection area, the accuracy of contour detection is improved, and the interference of the quartz ingot on contour detection during the injection production process is reduced;

[0089] The system is connected to a computer image processing server through an interface controller and has functions of real-time data recording, alarm, historical data query, and report output. This enables an alarm signal to be immediately issued when an abnormal quality of the quartz ingot is found, helping the operator to quickly handle the problem. In addition, all detection data can be recorded and stored according to time stamps, providing a reliable basis for later quality traceability and further enhancing the controllability and transparency of the production process;

[0090] Through the automated detection system, the present invention reduces human interference and ensures the accuracy and consistency of detection, especially in high-temperature and complex production environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 is a flowchart of an on-line quality detection method for quartz ingots based on improved YOLOv8 shown according to an exemplary embodiment;

[0092] Figure 2 is a network structure diagram of an on-line quality detection method for quartz ingots based on improved YOLOv8 shown according to an exemplary embodiment;

[0093] Figure 3 is a schematic diagram of an on-line quality detection system for quartz ingots based on improved YOLOv8 shown according to an exemplary embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0094] Now, the content of the present invention will be described with reference to exemplary embodiments. It should be understood that the described embodiments are only for enabling those of ordinary skill in the art to better understand and thus implement the content of the present invention, rather than implying any limitation to the scope of the present invention.

[0095] As used herein, the term "comprising" and its variants are to be construed as open-ended terms meaning "including but not limited to". The term "based on" is to be construed as "at least partially based on", and the terms "one embodiment" and "an embodiment" are to be construed as "at least one embodiment".

[0096] According to an embodiment of the present invention, Figure 1 is a flowchart of an on-line quality detection method for quartz ingots based on improved YOLOv8 shown according to an exemplary embodiment. As Figure 1 shown, to achieve the above object, an on-line quality detection method for quartz ingots based on improved YOLOv8 provided by the present invention includes:

[0097] Obtain the production image of the quartz ingot in the kiln;

[0098] Figure 2 is a network structure diagram of an on-line quality detection method for quartz ingots based on improved YOLOv8 shown according to an exemplary embodiment. As Figure 2 shown, input the production image of the quartz ingot into the hierarchical feature extraction backbone network to obtain image features at different levels;

[0099] Input the image features at different levels into the feature enhancement and fusion network to obtain the feature enhancement and fusion features;

[0100] Input the feature enhancement and fused features into a multi-scale feature pyramid network to obtain multi-scale detection results.

[0101] According to an embodiment of the present invention, the hierarchical feature extraction backbone network includes five ghost convolutions with a step size of 2 and a convolution kernel of 3×3, four cross-stage feature fusion modules, a pyramid pooling layer, and an attention mechanism layer;

[0102] Input the quartz ingot production image in the kiln into the first ghost convolution, reduce the image size by half, and extract features of 64 channels to obtain the first convolutional feature, where the formula is,

[0103]

[0104] where, represents the ghost convolution;

[0105] P1 represents the first convolutional feature;

[0106] X represents the quartz ingot production image in the kiln;

[0107] Input the first convolutional feature into the second ghost convolution and the first cross-stage feature fusion module in sequence to obtain the second convolutional feature. Use a 3×3 convolution kernel with a step size of 2 to perform convolution on the quartz ingot image of 320×320×64 after the first stage processing, reduce the image size by half, and extract features of 128 channels, and use the C2f module to further fuse features of different stages and enhance the feature expression ability, where the formula is,

[0108]

[0109] where, C2f 3 represents the cross-stage feature fusion module with a repeated stacking number of 3;

[0110] P2 represents the second convolutional feature;

[0111] Input the second convolutional feature into the third ghost convolution and the second cross-stage feature fusion module in sequence to obtain the third convolutional feature. Send the quartz ingot image of 160×160×128 after the second stage processing into a 3×3 convolution kernel with a step size of 2 to perform convolution, reduce the image size by half, and extract features of 256 channels, and use the C2f module to strengthen feature fusion, where the formula is,

[0112]

[0113] where, C2f 6 represents the cross-stage feature fusion module with a repeated stacking number of 6;

[0114] P3 represents the third convolutional feature;

[0115] The third convolutional feature is sequentially input into the fourth ghost convolution and the third cross-stage feature fusion module to obtain the fourth convolutional feature. The processed quartz ingot image of 80×80×256 at the 3rd stage is sent to a 3×3 convolutional kernel with a stride of 2, and the image size is further reduced to 40×40 while the feature map dimension increases to 512, and the C2f module is used to continue feature fusion. The formula is as follows.

[0116]

[0117] where, P4 represents the fourth convolutional feature;

[0118] The fourth convolutional feature is sequentially input into the fifth ghost convolution, the fourth cross-stage feature fusion module, the pyramid pooling layer, and the attention mechanism layer. The processed quartz ingot image of 40×40×512 at the 4th stage is sent to a 3×3 convolutional kernel with a stride of 2, and the image size is further reduced to 20×20 while the feature map dimension increases to 1024, and then further feature fusion is performed through the C2f module. Subsequently, the CBAM attention mechanism is used to enhance the feature expression ability, and the SPPF pyramid pooling module is used to extract multi-scale information to obtain the fifth convolutional feature. The formula is as follows.

[0119]

[0120] where, P5 represents the fifth convolutional feature;

[0121] CBAM represents the attention mechanism layer;

[0122] SPPF represents the pyramid pooling layer;

[0123] C2f 3 represents the cross-stage feature fusion module with a repeated stacking number of 3.

[0124] According to an embodiment of the present invention, for the cross-stage feature fusion module, its calculation process C2f n (X) can be expressed as:

[0125] For the input feature map, the number of channels is first expanded through a 1×1 convolution to obtain the expanded feature. The formula is as follows.

[0126]

[0127] where, O represents the feature map;

[0128] O represents the expanded feature;

[0129] Conv 1×1 represents the 1×1 convolution;

[0130] Indicates that the element type in the tensor is a real number;

[0131] H represents the height of the tensor space;

[0132] W represents the width of the tensor space;

[0133] C′ represents the number of channels for convolution expansion, C′ = [e·C out , where e is the expansion coefficient;

[0134] The expanded features are evenly divided into two parts along the channel dimension, the first channel part S1 and the second channel part S2. Perform the Bottleneck operation n times on the second channel part, where the formula is,

[0135]

[0136] Bottleneck(x) = Conv 3×3 (SiLU(Conv 3×3 (x)))+x;

[0137] Among them, x represents the input feature of the Bottleneck operation;

[0138] Conv 3×3 Represents a 3×3 convolution;

[0139] SiLU represents the activation function;

[0140] Represents the i-th operation of the second channel part;

[0141] Represents the (i - 1)-th operation of the second channel part;

[0142] Concatenate the first channel part with the output of the Bottleneck operation and map it to the target number of channels through a 1×1 convolution to obtain the output of the cross-stage feature fusion module, where the formula is,

[0143]

[0144] The CBAM attention mechanism includes a channel-spatial dual attention mechanism, where the formula is,

[0145]

[0146] Among them, X1 represents the feature input of this operation mechanism;

[0147] X C Represents the intermediate feature output;

[0148] σ represents the sigmoid activation function;

[0149] MLP represents max pooling;

[0150] GAP represents global average;

[0151] represents channel stacking;

[0152] X s represents the output of the CBAM mechanism;

[0153] GhostConv 7×7 represents the ghost convolution with a convolution of 7×7.

[0154] According to an embodiment of the present invention, the feature enhancement and fusion network includes an upsampling fusion layer and a downsampling fusion layer;

[0155] The upsampling fusion layer includes three upsampling fusion modules. Each upsampling fusion module has the same architecture and is composed of a bilinear upsampling operation layer, a fusion layer, and a cross-stage feature fusion module;

[0156] Input the fifth convolution feature and the fourth convolution feature into the first upsampling fusion module to obtain the first fusion feature, where the formula is,

[0157]

[0158] where, Q1 represents the first fusion feature;

[0159] Concat represents the channel fusion mechanism;

[0160] represents the bilinear upsampling operation layer;

[0161] Input the first fusion feature and the third convolution feature into the second upsampling fusion module to obtain the second fusion feature, where the formula is,

[0162]

[0163] where, Q2 represents the second fusion feature;

[0164] Input the second fusion feature and the second convolution feature into the third upsampling fusion module to obtain the third fusion feature, where the formula is,

[0165]

[0166] where, Q3 represents the third fusion feature.

[0167] According to an embodiment of the present invention, the downsampling fusion layer includes three downsampling fusion modules, each of which has the same architecture and consists of a downsampling operation layer, a fusion layer, and a cross-stage feature fusion module;

[0168] Input the second fusion feature and the third fusion feature into the first downsampling fusion module to obtain a fourth fusion feature, where the formula is

[0169]

[0170] where Q4 represents the fourth fusion feature;

[0171] represents the downsampling operation layer;

[0172] Input the fourth fusion feature and the first fusion feature into the second downsampling fusion module to obtain a fifth fusion feature, where the formula is

[0173]

[0174] where Q5 represents the fifth fusion feature;

[0175] Input the fifth fusion feature and the fifth convolution feature into the third downsampling fusion module to obtain a sixth fusion feature, where the formula is

[0176]

[0177] where Q6 represents the sixth fusion feature.

[0178] According to an embodiment of the present invention, the multi-scale feature pyramid network includes feature channel superposition, classification convolution, and regularization convolution;

[0179] Input the third fusion feature, the fourth fusion feature, the fifth fusion feature, and the sixth fusion feature into the multi-scale feature pyramid network respectively to obtain detection results of different scales, where the formula is

[0180]

[0181] where M represents the multi-scale feature pyramid network;

[0182] Detect represents the detection results of different scales;

[0183] represents the classification convolution;

[0184] Q k represents different fusion features;

[0185] represents the regularization convolution;

[0186] Indicates the superposition of feature channels;

[0187] H k Indicates the height of the feature map at scale k;

[0188] W k Indicates the width of the feature map at scale k;

[0189] c represents the object confidence;

[0190] (x, y, w, h) represents the bounding box coordinate information;

[0191] C represents the class probability distribution.

[0192] According to an embodiment of the present invention, the bounding box coordinates of the bubbles, nodules and the surface of the quartz ingot are obtained based on the multi-scale detection results;

[0193] The detection area of each class is binarized according to the bounding box coordinates to remove background noise;

[0194] The Gaussian filter smoothing technique is used to denoise the image and the outer contours of the bubbles, nodules and the surface of the quartz ingot are extracted based on the sub-pixel contour detection technique, and the height of the ingot surface, the hump ratio and the time, position and size of the bubbles and nodules are calculated.

[0195] According to an embodiment of the present invention, the video data of the on-line production of the quartz ingot in the kiln is collected in real time by an industrial camera installed on the observation port of the quartz ingot manufacturing furnace. The collection time is 480 minutes. The video is frame-extracted to obtain images (one image is extracted every 30 frames). The images obtained by frame extraction are data-labeled, and the coordinate areas and categories of the quartz ingot surface, bubbles and nodules in the images are recorded to make a data set of the quartz ingot surface, bubbles and nodules. The training set, validation set and test set are divided according to the data set. The Copy-Paste data augmentation method is used to expand the data set to help enhance the training samples of the two rare categories of nodules and bubbles, and the structure of the original labels is not changed, ensuring that the sample features are fully extracted during the model training process, while ensuring the generalization ability of the model, preventing overfitting, so that the model maintains a high detection accuracy and detection speed in the complex kiln production background. The training effect of the model is evaluated by indicators such as MAP (mean Average Precision), precision and recall, and the best trained result model is left. When the input quartz ingot image in grayscale format is copied and expanded to 3 channels before entering the model, and then sent into the trained improved YOLOv8 model.

[0196] Moreover, to achieve the above-mentioned invention purpose, the present invention also provides a quartz ingot on-line quality detection system based on the improved YOLOv8, Figure 3Schematic diagram of an online quality inspection system for quartz ingots based on improved YOLOv8 shown according to an exemplary embodiment, as Figure 3 shown, an online quality inspection system for quartz ingots based on improved YOLOv8 in the present invention includes:

[0197] A quartz ingot manufacturing furnace for producing quartz ingots;

[0198] An imaging detection device is arranged on the observation port of the quartz ingot manufacturing furnace to collect online production data of the quartz ingot in the quartz ingot manufacturing furnace in real time;

[0199] A computer image processing device is used to receive the production data of the imaging detection device and analyze the production data;

[0200] A detection interface is connected to the computer image processing device to display the detection result in real time;

[0201] An interface controller is connected to the computer image processing device to obtain the detection result in real time.

[0202] According to one aspect of the present invention, the interface controller includes a communication module, an alarm module and a data recording module;

[0203] The communication module is used for data transmission with the computer image processing device;

[0204] The alarm module triggers an audible and visual alarm when the quality of the quartz ingot is detected to be abnormal, reminding the operator to handle it in time;

[0205] The data recording module records the detection data and time stamps and generates a data log file for later quality analysis and traceability.

[0206] Those of ordinary skill in the art can realize that the modules and algorithm steps described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0207] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the devices and equipment described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0208] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or modules can be in electrical, mechanical or other forms.

[0209] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the embodiments of the present invention.

[0210] In addition, each functional module in the embodiments of the present invention can be integrated into a processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0211] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or this part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method for sending / receiving energy-saving signals in each embodiment of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical discs that can store program codes.

[0212] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features with similar functions disclosed in the present application.

[0213] It should be understood that the magnitudes of the serial numbers of the steps in the summary of the invention and the embodiments of the present invention do not absolutely indicate the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

Claims

1. An on-line quality inspection method for quartz weights based on improved YOLOv8, characterized in that, Including: Obtain the production image of quartz ingots in the kiln; Input the production image of quartz ingots into the hierarchical feature extraction backbone network to obtain image features at different levels; Input the image features at different levels into the feature enhancement and fusion network to obtain feature enhancement and fusion features; Input the feature enhancement and fusion features into the multi-scale feature pyramid network to obtain multi-scale detection results.

2. The on-line quality inspection method for quartz weights based on improved YOLOv8 according to claim 1, characterized in that The hierarchical feature extraction backbone network includes five Ghost convolutions with a step of 2 and a convolution kernel of 3×3, four cross-stage feature fusion modules, a pyramid pooling layer, and an attention mechanism layer; Input the production image of quartz ingots in the kiln into the first Ghost convolution to obtain the first convolution feature, where the formula is, Among them, represents ghost convolution; P1 represents the first convolution feature; X represents the production image of quartz ingots in the kiln; Input the first convolution feature into the second Ghost convolution and the first cross-stage feature fusion module in sequence to obtain the second convolution feature, where the formula is, Among them, C2f 3 represents the cross-stage feature fusion module with a repeated stacking times of 3; P2 represents the second convolution feature; Input the second convolution feature into the third Ghost convolution and the second cross-stage feature fusion module in sequence to obtain the third convolution feature, where the formula is, Among them, C2f 6 represents the cross-stage feature fusion module with a repeated stacking times of 6; P3 represents the third convolution feature; Input the third convolution feature into the fourth Ghost convolution and the third cross-stage feature fusion module in sequence to obtain the fourth convolution feature, where the formula is, where, P4 represents the fourth convolution feature; Input the fourth convolution feature into the fifth Ghost convolution, the fourth cross-stage feature fusion module, the pyramid pooling layer, and the attention mechanism layer in sequence to obtain the fifth convolution feature, where the formula is, where, P5 represents the fifth convolution feature; CBAM represents the attention mechanism layer; SPPF represents the pyramid pooling layer; C2f 3 Indicates a cross-stage feature fusion module with a repeated stacking times of 3.

3. The on-line quality inspection method of quartz blanks based on improved YOLOv8 according to claim 2, characterized in that, The feature enhancement and fusion network includes an upsampling fusion layer and a downsampling fusion layer; The upsampling fusion layer includes three upsampling fusion modules, each with the same architecture, consisting of a bilinear upsampling operation layer, a fusion layer, and a cross-stage feature fusion module; Input the fifth convolution feature and the fourth convolution feature into the first upsampling fusion module to obtain the first fusion feature, where the formula is, where, Q1 represents the first fusion feature; Concat represents the channel fusion mechanism; Indicates a bilinear upsampling operation layer; Input the first fusion feature and the third convolution feature into the second upsampling fusion module to obtain the second fusion feature, where the formula is, where, Q2 represents the second fusion feature; Input the second fusion feature and the second convolution feature into the third upsampling fusion module to obtain the third fusion feature, where the formula is, where, Q3 represents the third fusion feature.

4. The on-line quality inspection method of quartz weights based on improved YOLOv8 according to claim 3, characterized in that, The downsampling fusion layer includes three downsampling fusion modules, each with the same architecture, consisting of a downsampling operation layer, a fusion layer, and a cross-stage feature fusion module; Input the second fusion feature and the third fusion feature into the first downsampling fusion module to obtain the fourth fusion feature, where the formula is, where, Q4 represents the fourth fusion feature; Indicates the downsampling operation layer; Input the fourth fusion feature and the first fusion feature into the second downsampling fusion module to obtain the fifth fusion feature, where the formula is, where, Q5 represents the fifth fusion feature; Input the fifth fusion feature and the fifth convolutional feature into the third downsampling fusion module to obtain the sixth fusion feature, where the formula is where Q6 represents the sixth fusion feature.

5. The on-line quality inspection method for quartz weights based on improved YOLOv8 according to claim 4, characterized in that, The multi-scale feature pyramid network includes feature channel stacking, classification convolution, and regular convolution; Input the third fusion feature, the fourth fusion feature, the fifth fusion feature, and the sixth fusion feature into the multi-scale feature pyramid network respectively to obtain detection results at different scales, where the formula is where M represents the multi-scale feature pyramid network; Detect represents the detection results at different scales; Indicates classification convolution; Q k represent different fusion features; Indicates regular convolution; Indicates the superposition of feature channels; H k Denotes the height of the feature map at scale k; W k represents the width of the feature map at scale k; c represents the object confidence; (x, y, w, h) represents the bounding box coordinate information; C represents the class probability distribution.

6. The on-line quality inspection method for quartz weights based on improved YOLOv8 according to claim 5, characterized in that Based on the multi-scale detection results, obtain the bounding box coordinates of the bubbles, nodules, and the surface of the quartz ingot; Perform binary processing on the detection regions of each class according to the bounding box coordinates to remove background noise; Use Gaussian filter smoothing technology to denoise the image and extract the outer contours of the bubbles, nodules, and the surface of the quartz ingot based on the sub-pixel contour detection technology, and calculate the height of the ingot surface, the hump ratio, and the time, position, and size of the appearance of bubbles and nodules.

7. An on-line quality inspection system for quartz weights based on the improved YOLOv8, which applies the on-line quality inspection method for quartz weights based on the improved YOLOv8 according to any one of claims 1 to 6, is characterized in that, It includes: A quartz ingot manufacturing furnace for producing quartz ingots; An imaging detection device is set on the observation port of the quartz ingot manufacturing furnace to collect the online production data of the quartz ingot in the quartz ingot manufacturing furnace in real time; A computer image processing device is used to receive the production data of the imaging detection device and analyze the production data; A detection interface is connected to the computer image processing device to display the detection results in real time; An interface controller is connected to the computer image processing device to obtain the detection results in real time.

8. The on-line quality inspection system for quartz weights based on the improved YOLOv8 according to claim 7, wherein, The interface controller includes a communication module, an alarm module, and a data recording module; The communication module is used for data transmission with the computer image processing device; The alarm module triggers an audible and visual alarm when the quality of the quartz ingot is detected to be abnormal, reminding the operator to deal with it in time; The data recording module records the detection data and time stamps and generates a data log file for later quality analysis and traceability.

Images