Aluminum foil sealing quality detection system

Abstract

The invention relates to the technical field of packaging quality detection, and discloses an aluminum foil sealing quality detection system, which comprises a transmission module, a detection module and a control module, the thermal imaging acquisition module is used for acquiring temperature distribution data of a sealing area on the to-be-detected container; the image processing module is used for analyzing the temperature distribution data based on an image recognition algorithm and extracting defect characteristics of the aluminum foil seal; the temperature calibration module is used for dynamically calibrating the thermal imaging acquisition module according to preset parameters; the sorting module is used for removing defective products from the conveying module according to the extracted defect features; and the intelligent decision module is used for integrating the detection data and the production parameters, generating process optimization suggestions and outputting regulation and control instructions. According to the system and the method, high-precision real-time detection and classification of aluminum foil sealing defects are realized, and meanwhile, the automation level, the detection stability and the quality control efficiency of a production line are improved through dynamic sorting and process optimization suggestion output.

Description

Technical Field

[0001] The present application relates to the technical field of packaging quality detection, and in particular to an aluminum foil sealing quality detection system. Background Art

[0002] With the increasing demand for both safety and efficiency in the food and pharmaceutical packaging industries, aluminum foil seal quality inspection has become a critical step in production lines. Traditional aluminum foil seal quality inspection relies on manual visual inspection, which is labor-intensive, inefficient, and prone to missed or false detections due to fatigue, making it difficult to meet the inspection requirements of modern high-speed production lines.

[0003] Existing inspection equipment mostly uses visible light image recognition technology, which has limited accuracy in identifying minor defects (such as loose seals, flipped aluminum foil, and missing aluminum foil). It is particularly effective in detecting hidden defects (such as poor internal seals). For example, traditional methods have difficulty capturing the tiny temperature differences between the aluminum foil and the bottle mouth using visible light, resulting in a high rate of missed detections. Furthermore, traditional inspection equipment is easily affected by factors such as ambient lighting and bottle color, resulting in reduced detection stability and affecting production cycle time. For example, dark bottles or reflective surfaces can significantly reduce the contrast of visible light images, thereby reducing the accuracy of defect identification.

[0004] Therefore, existing aluminum foil seal quality inspection methods have problems with low defect recognition accuracy and environmental adaptability. Therefore, a new inspection technology is urgently needed to achieve comprehensive control of aluminum foil seal quality. Summary of the Invention

[0005] The purpose of this application is to provide an aluminum foil sealing quality detection system to solve the technical problems raised in the above background technology.

[0006] To achieve the above objectives, the present application discloses the following technical solution: an aluminum foil seal quality detection system, comprising:

[0007] The conveying module is configured to: convey the containers to be inspected to the inspection area in sequence;

[0008] The thermal imaging acquisition module is configured to: acquire temperature distribution data of the sealing area on the container to be inspected;

[0009] An image processing module is configured to: analyze the temperature distribution data based on an image recognition algorithm to extract defect features of the aluminum foil seal;

[0010] a temperature calibration module configured to dynamically calibrate the thermal imaging acquisition module according to preset parameters to distinguish different types of defects;

[0011] A sorting module is configured to remove defective products from the conveying module according to the extracted defect characteristics;

[0012] The intelligent decision-making module is configured to integrate detection data and production parameters, generate process optimization suggestions and output control instructions.

[0013] Preferably, the sorting module includes a pneumatic push rod and a photoelectric sensor;

[0014] The photoelectric sensor is configured to: detect the position of the container;

[0015] The pneumatic push rod is configured to remove defective products on the conveying module according to the detection result of the container position.

[0016] Preferably, the conveying module operates at a speed of 350 bottles per minute.

[0017] Preferably, the thermal imaging acquisition module includes an infrared thermal imager.

[0018] Preferably, the temperature calibration module comprises:

[0019] A multi-point temperature calibration unit is configured to: perform multi-point temperature calibration on the infrared thermal imager before detection;

[0020] The dynamic compensation unit is configured to compensate for thermal imaging distortion caused by container movement in real time during the detection process.

[0021] Preferably, the real-time compensation of thermal imaging distortion caused by container movement during the detection process specifically includes:

[0022] Motion trajectory modeling: The motion velocity v and acceleration a of the container are collected by photoelectric sensors to establish a motion trajectory function:

[0023]

[0024] Thermal distortion compensation: The thermal imaging angle after dynamic compensation is calculated based on the thermal imaging distortion caused by container movement. The calculation formula is:

[0025]

[0026] Among them, θ raw is the original thermal imaging angle, is the instantaneous velocity of the container at time t, k is the heat conduction correction coefficient, is the temperature gradient vector.

[0027] Preferably, the image processing module includes:

[0028] The data analysis unit is configured to: extract defect features of the aluminum foil seal;

[0029] a defect classification unit configured to: analyze the defect type of the container according to the defect characteristics, wherein the defect types include a non-aluminum foil defect, a loose sealing defect, an aluminum foil flipping defect, and a minor poor sealing defect;

[0030] The defect localization unit is configured to accurately locate the position of the defect in the container sealing area through a bounding box annotation function.

[0031] Preferably, the intelligent decision-making module includes:

[0032] A data storage unit is configured to: store historical inspection data, defect types and corresponding process parameters;

[0033] A trend analysis unit is configured to: analyze the correlation between defect types and production process parameters based on a machine learning algorithm and generate process optimization suggestions;

[0034] The real-time early warning unit is configured to trigger a production line shutdown or process adjustment instruction when it detects that the defect rate exceeds a preset threshold.

[0035] Preferably, the analysis of the correlation between defect types and production process parameters based on a machine learning algorithm to generate process optimization suggestions specifically includes:

[0036] Data preprocessing: Construct the historical detection data D and process parameters P into a feature matrix:

[0037] X=[D1,D2,...,D n ; P1, P2, ..., P m ]

[0038] Model training: Using an integrated model of random forest and long short-term memory network, we optimize parameters through cross-validation and output the correlation function between defect rate and process parameters:

[0039]

[0040] Among them, ω i is the model weight, φ i (P) is the output of the i-th basis function;

[0041] Dynamic control: Generate process parameter adjustment suggestions based on the correlation function to reduce the defect rate to the target threshold R target .

[0042] Preferably, the defect classification unit performs defect type analysis on the defects of the container based on a multidimensional feature fusion model, wherein the multidimensional feature fusion model includes:

[0043] Feature extraction layer: The local temperature gradient feature ΔT and the global temperature mean feature T of the thermal imaging image are extracted through the bounding box area generated by the YOLO algorithm avg ;

[0044] Multi-scale fusion layer: Constructs a multi-scale feature vector based on the preset defect type:

[0045]

[0046] Among them, α, β and γ are dynamic weight coefficients, is the time temperature change rate;

[0047] Classification decision layer: Classify the fused feature vector through the support vector machine model and output the confidence level and corresponding defect type.

[0048] Beneficial effects: The aluminum foil sealing quality inspection system of the present application, by integrating thermal imaging, image recognition algorithms and big data analysis technology, can detect whether the container bottle mouth is sealed properly without unscrewing the bottle cap, and realizes high-precision real-time detection and classification of aluminum foil sealing defects, effectively solving the problems of low efficiency of manual inspection, difficulty in identifying hidden defects, sensitivity to environmental interference and lagging process control. At the same time, through dynamic sorting and process optimization suggestion output, the automation level of the production line, detection stability and quality control efficiency are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 This is a structural block diagram of the aluminum foil seal quality detection system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following is a clear and complete description of the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] In this document, the term "comprising" is intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0053] This embodiment provides a Figure 1 The aluminum foil seal quality inspection system shown includes:

[0054] The conveying module is configured to: convey the containers to be inspected to the inspection area in sequence;

[0055] The thermal imaging acquisition module is configured to: acquire temperature distribution data of the sealing area on the container to be inspected;

[0056] An image processing module is configured to: analyze the temperature distribution data based on an image recognition algorithm to extract defect features of the aluminum foil seal;

[0057] a temperature calibration module configured to dynamically calibrate the thermal imaging acquisition module according to preset parameters to distinguish different types of defects;

[0058] A sorting module is configured to remove defective products from the conveying module according to the extracted defect characteristics;

[0059] The intelligent decision-making module is configured to integrate detection data and production parameters, generate process optimization suggestions and output control instructions.

[0060] The aluminum foil seal quality inspection system of this embodiment integrates thermal imaging, image recognition algorithms, and big data analysis technologies. It can detect whether the container bottle mouth is properly sealed without unscrewing the bottle cap, achieving high-precision real-time detection and classification of aluminum foil seal defects. It effectively solves problems such as low manual inspection efficiency, difficulty in identifying hidden defects, sensitivity to environmental interference, and lagging process control. At the same time, through dynamic sorting and process optimization suggestion output, the automation level of the production line, detection stability, and quality control efficiency are improved.

[0061] In one embodiment, the sorting module includes a pneumatic push rod and a photoelectric sensor;

[0062] The photoelectric sensor is configured to: detect the position of the container;

[0063] The pneumatic push rod is configured to remove defective products from the conveying module according to the detection result of the container position. It is feasible that this removal action is completed within 0.5 seconds.

[0064] It is feasible that the conveying module operates at a speed of 350 bottles per minute and is suitable for containers with a bottle mouth diameter ranging from 20 mm to 60 mm.

[0065] In one embodiment, the thermal imaging acquisition module includes an infrared thermal imager (such as the FOTRIC 626C or FLIR series) with a resolution of at least 640×480, a frame rate of at least 50 frames per second, and a detection wavelength range of 7.5μm-13μm to accommodate the thermal radiation characteristics of containers of different materials. This ensures clear and real-time thermal images, meeting the inspection requirements of high-speed production lines (e.g., 350 bottles per minute). The wide wavelength range design covers the thermal radiation characteristics of common container materials (glass, plastic, metal), avoiding detection errors caused by material differences.

[0066] In this way, by combining the thermal imaging acquisition module, the conveyor module, and the sorting module, the system supports the inspection of a wide range of container sizes (such as beverage bottles and pharmaceutical bottles) without hardware replacement, reducing equipment costs. Furthermore, the combination of high-speed thermal imagers and the high-speed response of the sorting module ensures production cycle times, reduces rejection delays, and ensures zero missed inspections of defective products.

[0067] It is feasible that the temperature calibration module includes:

[0068] A multi-point temperature calibration unit is configured to: perform multi-point temperature calibration on the infrared thermal imager before detection to eliminate the influence of ambient temperature difference on the detection result;

[0069] The dynamic compensation unit is configured to compensate for thermal imaging distortion caused by container movement in real time during the detection process.

[0070] The temperature calibration module, which is composed of a multi-point temperature calibration unit and a dynamic compensation unit, eliminates the influence of ambient temperature fluctuations on the test results through multi-point calibration (for example, when the workshop temperature difference is ±5°C, the accuracy can still be maintained at ±0.03°C).

[0071] Furthermore, the real-time compensation of thermal imaging distortion caused by container movement during the detection process specifically includes:

[0072] Motion trajectory modeling: The motion velocity v and acceleration a of the container are collected by photoelectric sensors to establish a motion trajectory function:

[0073]

[0074] Thermal distortion compensation: Calculate the thermal imaging angle after dynamic compensation based on the thermal imaging distortion caused by container movement. The purpose is to eliminate the thermal imaging distortion caused by container movement and restore the true thermal distribution data. The calculation formula is:

[0075]

[0076] Among them, θ raw is the original thermal imaging angle, is the instantaneous velocity of the container at time t, The purpose of this step is to quantify the position change of the container during the thermal imaging process, so as to determine the impact of motion on the thermal imaging. k is the heat conduction correction coefficient, which is used to adjust the weight of the correction term. It is obtained through experimental calibration. The calibration method can be:

[0077] Standard test: measures the actual amount of thermal imaging distortion in a scene with known motion speed and temperature gradient;

[0078] Fitting formula: The value of k is fitted through experimental data, specifically:

[0079]

[0080] Where Δθ measured is the actual thermal imaging distortion amount obtained through experimental measurement;

[0081] is the temperature gradient vector, which is calculated from the temperature difference between adjacent pixels. T i,j is the temperature value of the pixel in the i-th row and j-th column in the thermal imaging image, This reflects the spatial variation characteristics of heat distribution. The larger the temperature gradient in an area (such as the edge of the aluminum foil seal), the more significant the distortion effect.

[0082] By eliminating thermal imaging distortion, we can achieve a thermal imaging angle deviation of ≤0.5° on a high-speed production line of 350 bottles per minute, ensuring image stability. In addition, the dynamic compensation algorithm has a response time of <10ms, which meets the needs of high-speed inspection.

[0083] In one embodiment, the image processing module includes:

[0084] The data analysis unit is configured to: extract defect features of the aluminum foil seal;

[0085] a defect classification unit configured to: analyze the defect type of the container according to the defect characteristics, wherein the defect types include a non-aluminum foil defect, a loose sealing defect, an aluminum foil flipping defect, and a minor poor sealing defect;

[0086] The defect localization unit is configured to accurately locate the position of the defect in the container sealing area through a bounding box annotation function.

[0087] By means of the image processing module, the intelligent decision-making module includes:

[0088] A data storage unit is configured to: store historical inspection data, defect types and corresponding process parameters;

[0089] A trend analysis unit is configured to: analyze the correlation between defect types and production process parameters based on a machine learning algorithm and generate process optimization suggestions;

[0090] The real-time early warning unit is configured to trigger a production line shutdown or process adjustment instruction when it detects that the defect rate exceeds a preset threshold (such as >3%).

[0091] Furthermore, the machine learning algorithm is used to analyze the correlation between defect types and production process parameters and generate process optimization suggestions, specifically including:

[0092] Data preprocessing: Construct the historical detection data D and process parameters P into a feature matrix:

[0093] X=[D1,D2,...,D n ; P1, P2, ..., P m ]

[0094] Model training: Using an integrated model of random forest and long short-term memory network, we optimize parameters through cross-validation and output the correlation function between defect rate and process parameters:

[0095]

[0096] Among them, ω i is the model weight (optimized iteratively through training data), φ i (P) is the output of the i-th basis function, extracted by the intermediate layer features of random forest or LSTM;

[0097] Dynamic control: Generate process parameter adjustment suggestions based on the correlation function to reduce the defect rate to the target threshold R target , that is, by adjusting the process parameters P to make R defect Close to the target threshold R target .

[0098] In generating process optimization recommendations, the control recommendations output by the model reduce the defect rate to the target threshold (e.g., from 3% to 0.5%), and through the integrated model, take into account global characteristics and timing dependencies, and adapt to the dynamic changes of process parameters.

[0099] Furthermore, the defect classification unit performs defect type analysis on the defects of the container based on a multidimensional feature fusion model, wherein the multidimensional feature fusion model includes:

[0100] Feature extraction layer: The local temperature gradient feature ΔT and the global temperature mean feature T of the thermal imaging image are extracted through the bounding box area generated by the YOLO algorithmavg ;

[0101] Multi-scale fusion layer: Based on the preset defect type (such as no aluminum foil, loose seal), a multi-scale feature vector is constructed:

[0102]

[0103] Among them, α, β, and γ are dynamic weight coefficients, which are obtained by training historical defect samples; is the time temperature change rate, which is the rate at which the temperature in the thermal image changes over time (e.g., the temperature difference between consecutive frames);

[0104] Classification decision layer: The fused feature vector is classified using a support vector machine model (SVM model), and the confidence level and the corresponding defect type are output. The corresponding relationship between the confidence level and the defect type can be, but is not limited to, as shown in Table 1.

[0105] Table 1

[0106] Defect Type Confidence ρ No aluminum foil defects 0.9<ρ≤10 Loose sealing defect 0.6<ρ≤0.9 Aluminum foil flip defect 0.3<ρ≤0.6 Minor sealing defects 0≤ρ≤0.3

[0107] In this embodiment, α reflects the sensitivity of the local temperature gradient to defect classification (for example, a foil flip defect may be accompanied by a large temperature gradient). β reflects the impact of the global temperature mean on defect classification (for example, the absence of an aluminum foil defect may result in a higher overall temperature). γ reflects the effect of the time-temperature change rate on defect classification (for example, a minor sealing defect may be accompanied by a slow temperature rise). It should be noted that the dynamic weight coefficient is obtained through dynamic optimization of the training process of historical defect samples, and the specific acquisition steps are as follows:

[0108] Feature standardization: ΔT, T avg 、 Perform normalization (such as Z-score standardization) to eliminate dimensional differences;

[0109] Kernel function selection: select kernel functions (such as RBF kernel and linear kernel) according to data distribution, and map features to high-dimensional space to improve classification performance;

[0110] Parameter optimization: Adjust SVM parameters (such as regularization coefficients and kernel function parameters) through cross-validation and Bayesian optimization;

[0111] It is feasible that the optimization of the dynamic weight coefficient is specifically achieved through the following steps:

[0112] Objective function: maximize the classification interval and minimize the classification error;

[0113] Optimization method: Use the gradient descent method to dynamically adjust α, β, and γ so that the fused feature vector can maximize the classification interval of different defect types.

[0114] As described above, the SVM model achieves a 98% or higher accuracy rate for defect identification, with confidence quantifying classification reliability (e.g., confidence level for no aluminum foil defects > 0.92). Furthermore, multi-dimensional feature fusion reduces the impact of environmental noise (e.g., bottle color changes) on classification.

[0115] In summary, the aluminum foil seal quality inspection system of this embodiment uses an infrared thermal imager to achieve non-contact temperature monitoring with a maximum accuracy of ±0.03°C. It can accurately capture temperature distribution differences in the sealing area (such as local overheating / undercooling) and directly identify small or hidden defects (such as poor internal sealing and aluminum foil flipping), overcoming the reliance of traditional visible light inspection on surface defects. In addition, infrared thermal imaging technology is not affected by ambient light, bottle color, etc., and combined with an automatic temperature calibration algorithm, it ensures the stability and consistency of the inspection results. Secondly, combined with the bounding box positioning and Multi-scale feature fusion enables high-confidence identification of defect types (such as absence of aluminum foil, loose seals, etc.) through the SVM classification model, significantly improving classification accuracy in complex scenarios. Furthermore, a real-time compensation algorithm based on the heat conduction model and container motion trajectory eliminates thermal imaging distortion caused by motion blur, ensuring the stability and reliability of thermal images on high-speed production lines. Finally, a trend analysis unit correlates process parameters with defect rates to generate dynamic control suggestions (such as adjusting sealing temperature and pressure), achieving closed-loop management from defect detection to production optimization, reducing scrap rates and improving production efficiency.

[0116] In the embodiments provided herein, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, a microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a computer. Computer-readable storage media can include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer.

[0117] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. An aluminum foil sealing quality detection system, characterized in that: include: The conveying module is configured to: convey the containers to be inspected to the inspection area in sequence; The thermal imaging acquisition module is configured to: acquire temperature distribution data of the sealing area on the container to be inspected; an image processing module configured to: analyze the temperature distribution data based on an image recognition algorithm to extract defect features of the aluminum foil seal; a temperature calibration module configured to dynamically calibrate the thermal imaging acquisition module according to preset parameters to distinguish different types of defects; A sorting module is configured to remove defective products from the conveying module according to the extracted defect characteristics; The intelligent decision-making module is configured to integrate detection data and production parameters, generate process optimization suggestions and output control instructions.

2. The aluminum foil sealing quality detection system according to claim 1, characterized in that: The sorting module includes a pneumatic push rod and a photoelectric sensor; The photoelectric sensor is configured to: detect the position of the container; The pneumatic push rod is configured to remove defective products on the conveying module according to the detection result of the container position.

3. The aluminum foil sealing quality detection system according to any one of claims 1-2, characterized in that: The working speed of the conveying module is 350 bottles / minute.

4. The aluminum foil sealing quality detection system according to claim 1, characterized in that: The thermal imaging acquisition module includes an infrared thermal imager.

5. The aluminum foil sealing quality detection system according to claim 4, characterized in that: The temperature calibration module includes: A multi-point temperature calibration unit is configured to: perform multi-point temperature calibration on the infrared thermal imager before detection; The dynamic compensation unit is configured to compensate for thermal imaging distortion caused by container movement in real time during the detection process.

6. The aluminum foil sealing quality inspection system according to claim 5, characterized in that: The real-time compensation of thermal imaging distortion caused by container movement during the inspection process specifically includes: Motion trajectory modeling: The motion velocity v and acceleration a of the container are collected by photoelectric sensors to establish a motion trajectory function: Thermal distortion compensation: The thermal imaging angle after dynamic compensation is calculated based on the thermal imaging distortion caused by container movement. The calculation formula is: Among them, θ raw is the original thermal imaging angle, is the instantaneous velocity of the container at time t, k is the heat conduction correction coefficient, is the temperature gradient vector.

7. The aluminum foil sealing quality inspection system according to claim 1, characterized in that: The image processing module includes: The data analysis unit is configured to: extract defect features of the aluminum foil seal; a defect classification unit configured to: analyze the defect type of the container according to the defect characteristics, wherein the defect types include a non-aluminum foil defect, a loose sealing defect, an aluminum foil flipping defect, and a minor poor sealing defect; The defect localization unit is configured to accurately locate the position of the defect in the container sealing area through a bounding box annotation function.

8. The aluminum foil sealing quality detection system according to claim 7, characterized in that: The intelligent decision-making module includes: A data storage unit is configured to: store historical inspection data, defect types and corresponding process parameters; A trend analysis unit is configured to: analyze the correlation between defect types and production process parameters based on a machine learning algorithm and generate process optimization suggestions; The real-time early warning unit is configured to trigger a production line shutdown or process adjustment instruction when it detects that the defect rate exceeds a preset threshold.

9. The aluminum foil sealing quality detection system according to claim 8, characterized in that: The machine learning algorithm is used to analyze the correlation between defect types and production process parameters and generate process optimization suggestions, specifically including: Data preprocessing: Construct the historical detection data D and process parameters P into a feature matrix: X=[D1,D2,...,D n ;P1,P2,...,P m ] Model training: Using an integrated model of random forest and long short-term memory network, we optimize parameters through cross-validation and output the correlation function between defect rate and process parameters: Among them, ω i is the model weight, φ i (P) is the output of the i-th basis function; Dynamic control: Generate process parameter adjustment suggestions based on the correlation function to reduce the defect rate to the target threshold R target .

10. The aluminum foil sealing quality inspection system according to claim 7, characterized in that: The defect classification unit performs defect type analysis on the container defects based on a multi-dimensional feature fusion model, wherein the multi-dimensional feature fusion model includes: Feature extraction layer: The local temperature gradient feature ΔT and the global temperature mean feature T of the thermal imaging image are extracted through the bounding box area generated by the YOLO algorithm avg ; Multi-scale fusion layer: Constructs a multi-scale feature vector based on the preset defect type: Among them, α, β and γ are dynamic weight coefficients, is the time temperature change rate; Classification decision layer: Classify the fused feature vector through the support vector machine model (SVM model) and output the confidence level and the corresponding defect type.

Images