Corrugated carton damage detection method based on machine vision

Abstract

The invention relates to the technical field of corrugated carton detection, in particular to a corrugated carton damage detection method based on machine vision, which comprises the following steps of: 1, acquiring and preprocessing surface data of a corrugated carton, 2, extracting characteristics of the corrugated carton and constructing a model, and 3, detecting and classifying the damage of the corrugated carton. Step 4, feeding back a detection result of the corrugated carton; in the scheme, high precision and comprehensiveness of corrugated carton damage detection are realized through multi-modal data fusion, dynamic sequence analysis, feature extraction and modeling technologies; according to multi-modal data fusion, a visible light image, a depth image and an infrared image are comprehensively utilized, richer and complementary information is obtained, and possible missing damage of a single modal can be detected; dynamic sequence analysis captures the development and change process of damage, potential progressive damage is found in time, and an accurate and reliable basis is provided for quality evaluation of corrugated cartons.

Description

Technical Field

[0001] The present invention relates to the technical field of corrugated cardboard box detection, and particularly to a method for detecting damage to corrugated cardboard boxes based on machine vision. Background Art

[0002] In the fields of modern logistics and commodity packaging, corrugated cardboard boxes, as a widely used packaging material, their quality and integrity are crucial for protecting the safe transportation of goods. However, during the production, transportation, and storage of corrugated cardboard boxes, the boxes are often damaged due to various reasons.

[0003] In traditional detection methods, when using static image analysis, a single image of the corrugated paper is first obtained using a camera or scanner, and then preprocessing operations such as denoising and enhancing contrast are performed on the collected image to improve the image quality. Next, image processing algorithms are used to extract features such as the edges, shapes, and textures of the corrugated paper image. Finally, based on the extracted features and the feature standards of the preset intact corrugated paper, a comparison is made to determine whether there is damage. For example, discontinuous edges, abnormal shapes, or abnormal texture changes may be considered as damage.

[0004] However, the above method only analyzes a single image and cannot capture the changes and development of damage over time. It may be difficult to detect some small and gradually formed damages, and it is easily affected by factors such as lighting and shooting angle, resulting in inaccurate detection results. Therefore, the present invention proposes a method for detecting damage to corrugated cardboard boxes based on machine vision. Summary of the Invention

[0005] Technical problems to be solved: Analyzing a single image cannot capture the changes and development of damage over time. It may be difficult to detect some small and gradually formed damages, and it is easily affected by factors such as lighting and shooting angle, resulting in inaccurate detection results.

[0006] In view of the deficiencies of the prior art, the present invention provides a method for detecting damage to corrugated cardboard boxes based on machine vision, thereby solving the technical problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0008] A method for detecting damage to corrugated cardboard boxes based on machine vision includes the following steps:

[0009] Step 1: Install multiple calibrated industrial high-definition cameras at different angles and install ring-shaped fill lights around them. The brightness of the fill lights is controlled by an automatic dimming system to ensure that all sides of the corrugated cardboard box are completely photographed, the light is uniform, shadows and reflections are reduced, and the image contrast and clarity are guaranteed.

[0010] The collected images are grayscale processed to highlight features such as shape, edges, and contours, which is beneficial for subsequent damage detection, improves the operation speed, reduces costs, and assigns weights according to the importance of the RGB color channels to the human eye vision. The bilateral filtering algorithm is used to denoise the images, and the parameters are reasonably adjusted to remove noise while retaining edge information.

[0011] Step 2: Extract the features of the corrugated cardboard box and model construction. By accurately extracting effective features and constructing a suitable model, the damaged area and degree can be more accurately identified to reduce misjudgment and missed judgment, realizing automatic detection. The constructed model can quickly and automatically detect a large number of corrugated cardboard boxes, improving the detection efficiency and saving labor and time costs. It can adapt to various changes, adapt to the feature differences caused by factors such as different batch production processes and materials, enhance generality and robustness, predict the damage trend, and use historical data and the model to predict the emerging damage trend and take preventive measures in advance.

[0012] S21. Extract the texture features of the corrugated cardboard box, which play the roles of accurately positioning the damaged area, improving the detection sensitivity, distinguishing the damage types, eliminating interference, enhancing the detection reliability, realizing early detection, adapting to various scenarios, and assisting in quantitative evaluation. The mutation or abnormality of the texture features accurately indicates the damage location, discovers the subtle changes caused by minor damage to enhance the sensitivity. Different damage types have unique texture change patterns to help distinguish. The stable texture features exclude interference in complex backgrounds and focus on the damage.

[0013] Judge that the texture change caused by damage reflects the quality problem caused by damage, compare the texture differences between intact and damaged cardboard boxes to assist in determining the damage location and scope, and assist in analyzing the impact of damage on the texture of the cardboard box material, which helps to improve the accuracy and efficiency of damage detection. Through the gray-level co-occurrence matrix method, the texture is described by calculating the frequency of pixel pairs appearing in a specific direction and distance, and is calculated by the following formula:

[0014]

[0015] Among them, p(i,j) is the value of the element (i,j) in the gray-level co-occurrence matrix.

[0016] S22. Extract the edge features of the corrugated cardboard box

[0017] Judge that the edge discontinuity or abnormality caused by damage determines the damage location. By detecting the change of edge curvature to detect minor damage and comparing it with the edges of intact cardboard boxes to assist in judging the damage degree, improving the accuracy and precision of damage detection. It is calculated by the canny edge detection algorithm. First, perform Gaussian filtering to smooth the image, then calculate the gradient magnitude and direction, and obtain the edge through double-threshold processing and edge connection. The gradient magnitude M and direction θ can be calculated by the following formula:

[0018]

[0019] Among them, G x and G y are the gradients in the horizontal and vertical directions respectively.

[0020] S23. Establish a multimodal feature fusion model

[0021] By establishing a multimodal feature fusion model, more subtle and complex patterns can be captured, thereby improving the accuracy of detecting and classifying damaged corrugated cartons. A single modality may miss some key features due to its own limitations, and when a certain modality is interfered with or of poor quality, the information of other modalities can play a supplementary and corrective role, making the model more robust in various complex actual scenarios. Suppose the feature vectors extracted from visible light images, depth images, and infrared images are V (with dimension n u ), D (with dimension n d ), and I (with dimension n i ) respectively. First, these three feature vectors are concatenated to obtain the fused feature vector F: F = [V; D; I], where Then, the fused feature vector F is input into a fully connected layer for processing.

[0022] S24. Establish a dynamic sequence analysis model

[0023] It can capture the morphological change trend of the corrugated carton, and estimate the degree of damage to the small gaps of the carton over time by analyzing the subtle changes between consecutive frames, laying a foundation for the subsequent classification of the corrugated carton. It also combines multi-frame information for judgment to reduce the occurrence of misjudgment caused by the instantaneous blurring of a single-frame image.

[0024] Step 3: Detection and classification of damaged corrugated cartons

[0025] Through image feature comparison and analysis, accurately identify the damaged location, quantify the size, area, depth, etc. of the damage and precisely describe it, detect minor damage, distinguish non-damaged factors, reduce the probability of misjudgment and missed judgment, and summarize the image feature patterns of different damages for subsequent detection. Topological structure change detection accurately identifies the damaged area, reveals the influence range, detects complex and hidden damages, improves robustness, supports early warning, assists in understanding the damage mechanism, reduces the misjudgment rate, adapts to different types of damages, and captures the connection relationship of the surface elements of the carton to evaluate the damage according to the changes. In addition, set a "completeness index", and classify the damage according to the comparison of its operation result with the threshold.

[0026] Step 4: Feedback the detection results of the corrugated carton

[0027] Adopt the TCP / IP protocol, use the hash algorithm MD5 to generate identifiers and check codes, split the data using packet switching and add information such as source address, destination address, and sequence number, and display detailed data such as damage type, location, and degree on the display screen.

[0028] In one implementation, since the image is inevitably affected by noise interference during the acquisition process, which affects the subsequent detection accuracy, a bilateral filtering algorithm is used for noise reduction. This algorithm can retain the edge information of the image while removing noise, and its filtering formula is:

[0029]

[0030] Among them, I(x,y) is the pixel value of the original image at (x,y), and I filtered (x,y) is the pixel value of the filtered image at (x,y), w 8 is the spatial domain weight function, and w r is the range domain weight function.

[0031] In one implementation, in order to further highlight the damage features in the image, the method of histogram equalization is used to enhance the noise-reduced image. By adjusting the gray histogram of the image, the gray distribution of the image becomes more uniform, and the contrast of the image is improved. The specific operation is to re-distribute the gray values of the image according to a certain mapping relationship, so that the areas with relatively low contrast become clearer.

[0032] In one implementation, accurately identify the damage location by comparing the normal and abnormal image features, provide accurate descriptions of the size, area, depth, etc. of the damage, detect tiny damages that are not easily detectable, distinguish the image changes caused by non-damage factors and focus on the real damage features, reduce the probability of false positives and missed detections to ensure reliability, and summarize and classify the image feature patterns of different damages for subsequent detection.

[0033] In one implementation, the construction of the topological structure is used to accurately identify the damaged area, reveal the scope of the damage, detect complex and hidden damages, improve the robustness of detection, support early warning, assist in understanding the damage mechanism, reduce the misjudgment rate, and adapt to different types of damages. The topological structure captures the connection relationships of the elements on the surface of the cardboard box. When damaged, the connection changes to accurately locate the area. By observing the changes, it can be understood whether the damage is local or large-scale, which helps to evaluate the severity. For unobvious or internal damages, the changes provide key clues because internal damages also affect the overall connection pattern. It is not affected by the geometric deformation of the cardboard box appearance, the shooting angle, and the lighting conditions, and focuses on the structural changes to achieve more stable and reliable detection. When the damage first appears and causes minor changes in the topological structure, detection is carried out to achieve early warning and avoid expansion, which helps to analyze the occurrence and spread of the damage and provides information for improving the design and production process. Based on objective changes for judgment, it reduces misjudgments caused by subjective visual judgment or image noise. For different types of damages such as cracks, holes, and tears, the changes are all reflected, showing wide applicability.

[0034] In one implementation, it is classified according to the degree of damage. A "complete index I" is set. The larger the value of I, the more complete it is, and the smaller the value, the more severe the damage. Through empirical data, the range of I is set as follows: the excellent range is I≥0.9; the good range is 0.7≤I<0.9; the defective range is I<0.7. If the integrity index is lower than 0.7, the cost of repairing or remanufacturing the cardboard box will exceed the benefits brought by its continued use.

[0035] In one implementation, between 0.7 and 0.9, through some simple treatments or use under specific conditions, it still meets certain requirements and has certain economic benefits; if it is higher than 0.9, no additional treatment is required and it can be used directly. According to the result of the operation, it is compared with the set threshold for reclassification.

[0036] Beneficial effects compared with the prior art:

[0037] 1. In this solution, through multi-modal data fusion, dynamic sequence analysis, feature extraction, and modeling techniques, high-precision and comprehensiveness of corrugated cardboard box damage detection are achieved; multi-modal data fusion comprehensively utilizes visible light images, depth images, and infrared images to obtain richer and complementary information, and can detect damages that may be missed by a single modality; dynamic sequence analysis captures the development and change process of the damage and timely discovers potential progressive damages; the precise extraction of features such as the texture and edges of the corrugated cardboard box and the construction of multiple models greatly improve the detection ability of complex and minor damages, reduce the probability of misjudgment and missed judgment, and provide an accurate and reliable basis for the quality assessment of corrugated cardboard boxes.

[0038] 2. In this solution, through the precise detection and classification of corrugated cardboard box damage, the optimization of production processes and procedures and the effective control of costs are achieved; the problems found during the detection process can be promptly fed back to the manufacturer, and the manufacturer can accordingly adjust the production process, adopt more advanced manufacturing technologies, optimize material selection, standardize the operation process, effectively reduce the occurrence of damage at the source, improve product quality, and also reduce additional costs such as goods damage, repackaging, and transportation delays caused by cardboard box damage. At the same time, by reasonably classifying and setting "integrity indicators", resources can be more effectively allocated, production efficiency can be improved, and a significant increase in economic benefits can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above description is only an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and implement it in accordance with the content of the description, the following will describe in detail with reference to the preferred embodiments of the present invention and the accompanying drawings.

[0040] Figure 1 is a schematic diagram of the overall structure of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can be implemented in various different forms, so the present invention is not limited to the embodiments described below;

[0042] The technical solutions in the embodiments of the present application are to solve the problems in the above background technology, and the general idea is as follows:

[0043] Embodiment 1:

[0044] Please refer to Figure 1 shown. This embodiment introduces a method for detecting corrugated cardboard box damage based on machine vision, including the following steps:

[0045] Step 1. Acquisition and preprocessing of surface data of corrugated cardboard boxes

[0046] S11. Multi-angle image acquisition

[0047] By setting multiple industrial high-definition cameras at different angles, calibrating the industrial high-definition cameras to ensure that each surface of the corrugated cardboard box can be completely photographed, and installing a ring-shaped fill light around the cameras so that the light can evenly illuminate the surface of the corrugated cardboard box, reducing the generation of shadows and avoiding the occurrence of reflective or shadow areas that affect the image quality. The brightness of the fill light is controlled by an automatic dimming system and is controlled in real time according to the current light intensity, reducing the occurrence of color difference, image blurring, and contrast reduction, and ensuring that the collected images have stable contrast and clarity.

[0048] S12. Grayscale processing of the image

[0049] The collected images contain rich color information, which is not convenient for detecting the damage of corrugated paper. Therefore, the images are grayscale processed to reduce the interference of color information, highlight the features such as the shape, edges and contours of the images more prominently, facilitate the subsequent detection of corrugated paper damage, and at the same time, grayscaling the images may also improve the operation speed of the image processing algorithm and reduce the calculation cost. According to the importance of different color channels in the RGB color space to human eye vision, different weights are assigned, and the calculation formula is: Gray = 0.299R + 0.587G + 0.114B.

[0050] S13. Image denoising processing

[0051] Since the images are inevitably interfered by noise during the acquisition process, which affects the subsequent detection accuracy, the bilateral filtering algorithm is used for denoising processing. This algorithm can retain the edge information of the images while removing the noise, and its filtering formula is:

[0052]

[0053] where I(x, y) is the pixel value of the original image at (x, y), I filtered (x, y) is the pixel value of the filtered image at (x, y), w 8 is the spatial domain weight function, w r is the range domain weight function. In practical applications, according to the noise situation and detail requirements of the images, the parameters of the bilateral filtering algorithm, such as the spatial domain standard deviation and the range domain standard deviation, are reasonably adjusted to achieve the best denoising effect.

[0054] S14. Image enhancement

[0055] To further highlight the damage features in the images, the histogram equalization method is used to enhance the denoised images. By adjusting the grayscale histogram of the images, the grayscale distribution of the images becomes more uniform, and the contrast of the images is improved. The specific operation is to reassign the grayscale values of the images according to a certain mapping relationship, so that the areas with relatively low contrast become clearer.

[0056] Step 2. Extract the features of the corrugated cardboard box and model construction

[0057] S21. Extract the texture features of the corrugated cardboard box

[0058] The texture changes caused by damage can be judged to reflect the quality problems caused by damage, used to compare the texture differences between intact and damaged cardboard boxes, assist in determining the location and scope of damage, and assist in analyzing the impact of damage on the texture of the cardboard box material, which helps to improve the accuracy and efficiency of damage detection. Through the gray-level co-occurrence matrix method, the texture is described by calculating the frequency of pixel pairs appearing in a specific direction and distance, and can be calculated by the following formula:

[0059]

[0060] Among them, p(i, j) is the value of the element (i, j) in the gray-level co-occurrence matrix.

[0061] S22. Extract the edge features of the corrugated cardboard box

[0062] It can judge the edge discontinuity or abnormality caused by damage to determine the damage location. By detecting the change of edge curvature to detect subtle damage and comparing it with the edge of the intact cardboard box, it can assist in judging the degree of damage and improve the accuracy and precision of damage detection. It can be calculated by the way of the canny edge detection algorithm. First, perform Gaussian filtering to smooth the image, then calculate the gradient magnitude and direction, and obtain the edge through double-threshold processing and edge connection. The gradient magnitude M and direction θ can be calculated by the following formula:

[0063]

[0064] Among them, G x and G y are the gradients in the horizontal and vertical directions respectively.

[0065] S23. Establish a multi-modal feature fusion model

[0066] By establishing a multi-modal feature fusion model, it can capture more subtle and complex patterns, thereby improving the accuracy of damage detection and classification of the target corrugated cardboard box. A single modality may miss some key features due to its own limitations, and when a certain modality is interfered with or has poor quality, the information of other modalities can play a supplementary and corrective role, making the model more robust in various complex actual scenarios. Suppose the feature vectors extracted from visible light images, depth images, and infrared images are V (dimension n u ), D (dimension n d ), and I (dimension n i ) respectively. First, splice these three feature vectors to obtain the fused feature vector F: F = [V; D; I], where, Then, input the fused feature vector F into a fully connected layer for processing.

[0067] S24. Establish a dynamic sequence analysis model

[0068] It can capture the morphological change trend of the corrugated cardboard box, and estimate the degree of damage to the tiny gaps of the cardboard box over time by analyzing the subtle changes between consecutive frames, laying a foundation for the subsequent classification of the corrugated cardboard box. It can also combine multi-frame information for judgment to reduce the occurrence of misjudgment caused by the instantaneous blurring of a single-frame image during photographing;

[0069] Calculated by the optical flow algorithm. Assume that in two adjacent frames I(x, y, t) and I(x + Δx, y + Δy, t + Δt), the brightness is constant, i.e., I(x, y, t) = I(x + Δx, y + Δy, t + Δt). Then, expanding the right side using the Taylor series gives: I(x, y, t) = I(x, y, t) + I x Δx + I y Δy + I t Δt + ∈, where I x , I y and I t are the partial derivatives of the image in the x, y, and t directions respectively, and ∈ is the high-order term. Due to the constant brightness assumption, the high-order term is ignored, resulting in: I x Δx + I y Δy + I t Δt = 0 For multiple pixels within a small window, it can be represented in matrix form:

[0070]

[0071] By using the least squares method to solve this linear system, the optical flow estimation is obtained:

[0072]

[0073] S25. Establish the topological structure

[0074] Accurately describing the characteristics of the damaged area and its relationship with the surroundings helps in precise positioning, discovering the topological structure changes caused by damage so as to detect the damage in a timely manner, improving the detection accuracy by comparing the topological structure differences between intact and damaged corrugated papers, enhancing the detection ability by using topological features to identify complex and tiny damage patterns, providing a basis for assessing the damage degree to facilitate formulating repair strategies, analyzing the state of corrugated paper in a more systematic and comprehensive way to improve the detection efficiency and effect, and establishing through the results extracted from image features.

[0075] Step 3. Detection and classification of corrugated box damage

[0076] S31. Comparative analysis of image features

[0077] Accurately identify the damage location by comparing the image features of normal and abnormal ones, quantify the damage size, area, depth, etc. to provide a precise description, detect tiny damages that are not easily noticeable, distinguish the image changes caused by non-damage factors and focus on the real damage features, reduce the probability of false judgment and missed judgment to ensure reliability, and summarize and classify the image feature patterns of different damages for subsequent detection.

[0078] S32. Detection of topological structure changes

[0079] Precisely identify the damaged area, reveal the scope of the damage, detect complex and hidden damages, improve the robustness of detection, support early warning, assist in understanding the damage mechanism, reduce the false positive rate, and adapt to different types of damages. The topological structure can capture the element connection relationships on the surface of the cardboard box. When damaged, the changed connections can accurately locate the area. By observing the changes, it can be understood whether the damage is local or extensive, which helps to evaluate the severity. For unobvious or internal damages, the changes may provide key clues because internal damages may also affect the overall connection pattern. It is not affected by the geometric deformation of the cardboard box appearance, the shooting angle, and the lighting conditions, focusing on the structural changes for a more stable and reliable detection. It can detect when the damage first appears, resulting in minor changes in the topological structure, to achieve early warning and avoid expansion. It helps to analyze the occurrence and spread of the damage, providing information for improving the design and production process. Based on objective changes for judgment, it reduces misjudgments caused by subjective visual judgment or image noise. For different types of damages such as cracks, holes, and tears, the changes can all be reflected, having wide applicability.

[0080] S33. Classify according to the degree of damage

[0081] Set a "complete index I", the larger its value indicates the more complete, and the smaller it indicates the more serious the damage. Through empirical data, the range of I is set as follows: the excellent range is I≥0.9; the good range is: 0.7≤I<0.9; the defective range is: I<0.7. If the integrity index is lower than 0.7, the cost of repairing or remanufacturing the cardboard box may exceed the benefits brought by its continued use; while between 0.7 and 0.9, it may still meet certain requirements through some simple treatments or use under specific conditions, having certain economic benefits; higher than 0.9 requires no additional treatment and can be used directly and normally. Through Classify again by comparing the operation result with the set threshold.

[0082] Step Four. Feedback the detection results of the corrugated cardboard box

[0083] S41. When detecting unqualified corrugated cardboard boxes, in order to ensure that relevant information can be transmitted to the display screen timely, accurately, and completely, the CP / IP protocol data transmission protocol will be adopted; by using the hash algorithm:

[0084] MD5(Message-Digest Algorithm5): MD5(message)=128-bit hash value

[0085] Generate unique identifiers and verification codes to ensure the accuracy and integrity of data. Then, use packet switching to divide the data into multiple data packets, and add information such as source address, destination address, and sequence number to each data packet. Display the detailed data of unqualified corrugated cartons on the display screen in a timely manner, including damage types, locations, degrees, etc., so that relevant personnel can quickly obtain key information and take corresponding measures.

[0086] Finally, it should be noted that: Obviously, the above embodiments are merely examples given to clearly illustrate the present invention, rather than limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation manners here. And the obvious changes or modifications derived therefrom still fall within the protection scope of the present invention.

Claims

1. A method for detecting damage of corrugated paperboard based on machine vision, characterized in that: The following steps are involved: Step 1: Set up multiple calibrated industrial high-definition cameras at different angles and install ring-shaped fill lights around them. The brightness of the fill lights is controlled by the automatic dimming system to fully capture all sides of the corrugated box. Grayscale the collected images to highlight features such as shape, edge and contour, assign weights according to the importance of RGB color channels to human vision, and then use the bilateral filtering algorithm to reduce image noise, reasonably adjust parameters, remove noise while retaining edge information; Step 2: Extract the features of corrugated boxes and build models. By accurately extracting effective features and building appropriate models, the damaged area and degree can be more accurately identified, misjudgment and missed judgment can be reduced, and automated detection can be achieved. The constructed model can quickly and automatically detect a large number of corrugated boxes, improve detection efficiency, save manpower and time costs, adapt to various changes, adapt to the feature differences caused by factors such as different batches of production processes and materials, enhance versatility and robustness, and predict damage trends; S21. Extract the texture features of corrugated boxes. The mutation or abnormality of texture features can accurately indicate the location of damage, and can detect subtle changes caused by small damage to enhance sensitivity. Stable texture features can eliminate interference and focus on damage in complex backgrounds. Determine the texture change caused by damage, reflect the quality problems caused by damage, and compare the texture difference between intact and damaged cartons to assist in determining the location and range of damage and assist in analyzing the impact of damage on the texture of the carton material. The texture is described by calculating the frequency of pixel pairs in a specific direction and distance through the gray level co-occurrence matrix, and the calculation is performed using the following formula: Where p(i,j) is the value of element (i,j) in the gray-level co-occurrence matrix; S22. Extract edge features of corrugated boxes Determine the edge discontinuity or abnormality caused by damage to determine the damage location. Detect slight damage by changing the edge curvature and compare it with the intact carton edge to assist in judging the degree of damage. Improve the accuracy and precision of damage detection. Then calculate it by the canny edge detection algorithm. First, use Gaussian filtering to smooth the image, then calculate the gradient amplitude and direction, and obtain the edge through double threshold processing and edge connection. The gradient amplitude M and direction θ are calculated by the following formula: Among them, G x and G y are the gradients in the horizontal and vertical directions respectively; S23. Establish a multimodal feature fusion model By establishing a multimodal feature fusion model, we can capture more subtle and complex patterns, thereby improving the accuracy of damage detection and classification of target corrugated boxes, making the model more robust in various complex practical scenarios. Assume that the feature vectors extracted from the visible light image, depth image, and infrared image are V (dimension n u ), D (dimension is n d ) and I (dimension n i ), first concatenate these three feature vectors to obtain the fused feature vector F: F = [V; D; I]; Where V = [u1,u2,…,u nu ],D=[d1,d2,…,d nu ],I=[i1,i2,…,i nu ], then the fused feature vector F is input into a fully connected layer for processing; S24. Establishing a dynamic sequence analysis model Capture the morphological change trend of corrugated boxes, and infer the degree of damage of tiny gaps in boxes over time by analyzing the subtle changes between consecutive frames, laying the foundation for subsequent classification of corrugated boxes, and making judgments based on multi-frame information; Step 3: Corrugated box damage detection and classification Through image feature comparison and analysis, it can accurately identify the damage location, quantify the damage size, area, depth, etc. and accurately describe it, detect minor damage, distinguish non-damage factors, reduce the probability of misjudgment and missed judgment, summarize the image feature patterns of different damage for subsequent detection, and accurately identify the damaged area through topological structure change detection. In addition, set the "complete index" and classify the damage according to the comparison of its calculation results with the threshold; Step 4: Feedback on the test results of corrugated boxes When unqualified corrugated boxes are detected, the TCP / IP protocol will be used, and the hash algorithm MD5 will be used to generate identification and verification codes. The data will be segmented using packet switching and information such as source address, destination address and serial number will be added. Detailed data such as damage type, location, and degree will be displayed on the display screen.

2. A method for detecting damage of corrugated paperboard based on machine vision as claimed in claim 1, characterized in that: The bilateral filtering algorithm is used for noise reduction. The filtering formula of this algorithm is as follows: Among them, I(x,y) is the pixel value of the original image at (x,y), I filtered (x, y) is the pixel value of the filtered image at (x, y), w8 is the spatial domain weight function, w r is the range weight function.

3. The method for detecting damage of corrugated paper boxes based on machine vision according to claim 1, characterized in that: In order to further highlight the damaged features in the image, the histogram equalization method is used to enhance the denoised image by adjusting the grayscale histogram of the image.

4. The method for detecting damage of corrugated paperboard based on machine vision according to claim 1, characterized in that: The damage location is accurately identified by comparing normal and abnormal image features, including quantifying the size, area, and depth of the damage to provide an accurate description, detecting subtle damage that is difficult to detect, and summarizing the image feature patterns of different damages to facilitate subsequent detection.

5. The method for detecting damage of corrugated paperboard based on machine vision according to claim 1, characterized in that: The construction of the topological structure can accurately identify the damaged area, reveal the impact range of the damage, detect complex and hidden damage, improve the robustness of detection, support early warning, and assist in understanding the damage mechanism. The topological structure can capture the connection relationship between elements on the surface of the carton, and detect when the damage just occurs and causes slight changes in the topological structure to achieve early warning and avoid expansion.

6. The method for detecting damage of corrugated paperboard based on machine vision according to claim 1, characterized in that: The classification is carried out according to the degree of damage, and an "integrity index I" is set. The larger the value, the more intact it is, and the smaller the value, the more serious the damage is. The range of I is set based on empirical data: excellent range I≥0.9; good range: 0.7≤I,<0.9; defective range: I<0.

7.

7. A method for detecting damage of corrugated paperboard based on machine vision as claimed in claim 6, characterized in that: When the integrity index is lower than 0.7, the cost of repairing or reproducing the carton will exceed the benefit of its continued use. When it is between 0.7 and 0.9, it can still be used with simple treatment and has certain economic benefits. When it is higher than 0.9, no additional treatment is required and it can be used normally. Reclassify based on the comparison of the calculation result with the set threshold.

Images