Real-time online detection method and system for fruit appearance defects

Abstract

The invention discloses a fruit appearance defect real-time online detection method and system, and relates to the technical field of image processing, and the system comprises an image capture module, a feature extraction and analysis module, a 3D reconstruction module, a dynamic segmentation module, a defect quantification and grading module, and a topology analysis and detection module. According to the real-time online detection method and system for the fruit appearance defects, the defects of a traditional method in the aspects of detection precision, classification fineness and environmental adaptability are overcome by combining a plurality of processes of multispectral image capture, hierarchical feature extraction, high-dimensional clustering analysis, 3D modeling, defect quantitative evaluation and the like. The system can accurately identify various defect types on the surface of the fruit in real time in a complex detection environment and quantify the severity of the defect types. Compared with the prior art, the system has the advantages that the accuracy, the stability and the real-time performance of defect detection are remarkably improved, an intelligent solution is provided for automatic sorting and quality control of fruits, and meanwhile, the labor cost is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a real-time online detection method for fruit appearance defects and a system thereof. Background Art

[0002] In modern fruit sorting and quality control, appearance defect detection is an important part of fruit quality assessment. Traditional detection methods mainly rely on artificial vision or automatic detection systems based on simple image processing algorithms. However, manual detection has the problems of low efficiency, high cost and strong subjectivity, and the existing automated detection technology still has many shortcomings under complex conditions. For example, a single spectral image is difficult to reveal subtle color changes or gloss anomalies on the surface of fruits; two-dimensional images have limited ability to detect surface concave and convex depth and complex morphological defects; background interference and lighting changes in dynamic detection environments also greatly affect the stability and accuracy of detection. In addition, existing defect classification methods are mostly based on simple threshold judgments, which cannot accurately distinguish different defect types and their severity, resulting in inaccurate sorting decisions. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a real-time online detection method and system for fruit appearance defects to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a real-time online detection method for fruit appearance defects, comprising:

[0005] Capture multispectral images of fruit surfaces;

[0006] Extract and analyze features of multispectral images to distinguish normal and abnormal areas on the fruit surface;

[0007] Generate a 3D model of the fruit surface based on multispectral images and zoom in on abnormal areas in detail;

[0008] Perform real-time defect detection and dynamic area segmentation, and perform defect feature detection and area segmentation on the fruit surface;

[0009] Conduct quantitative evaluation and classification modeling of defect characteristics and output defect classification results;

[0010] Conduct topological analysis of fruit skin texture and analyze abnormal textures in combination with defect grading results.

[0011] Further optimizing the technical solution, the capturing of the multispectral image of the fruit surface comprises:

[0012] Use a multispectral camera to capture image information from visible light to near-infrared bands on the fruit surface, identify color changes on the fruit surface, and detect texture and invisible defects on the fruit surface, including cracks and mold spots;

[0013] Optimize the light source angle, brightness, and polarization of the multispectral camera to reduce interference caused by surface reflections;

[0014] The contrast of different defect features on the fruit surface is enhanced by combining the spectral data of multispectral images in different bands.

[0015] To further optimize the technical solution, the feature extraction and analysis of the multispectral image includes:

[0016] Hierarchical feature extraction of multispectral images based on deep learning models;

[0017] High-dimensional clustering analysis is performed on the extracted features to distinguish normal and abnormal areas on the fruit surface.

[0018] To further optimize the technical solution, the layered feature extraction of the multispectral image includes:

[0019] The first layer is extraction, basic texture feature extraction;

[0020] The second layer is extraction, color anomaly and gloss change extraction;

[0021] The third layer is extraction, which extracts and analyzes the overall morphological characteristics and consistency of the fruit.

[0022] To further optimize the technical solution, the quantitative evaluation and classification modeling of defect characteristics include:

[0023] Key characteristic indicators of the quantitative evaluation system, including crack defect characteristics, spot defect characteristics, color abnormality characteristics, and concave or convex characteristics;

[0024] Construct a multidimensional classification model, which includes a comprehensive scoring function and scoring classification.

[0025] To further optimize this technical solution, the key characteristic indicators are:

[0026] Crack defect characteristics include:

[0027] length : The pixel distance of the crack in the image is calculated by edge detection method and converted into actual physical length;

[0028] width : Calculated by the width perpendicular to the main direction of the crack, taking the weighted value of the maximum width and the average width;

[0029] Distribution density : defined as the ratio of the total crack area to the fruit surface area;

[0030] Characteristics of spot defects include:

[0031] area : Directly calculate the total number of pixels in the spot area and convert it into the actual physical area;

[0032] density : defined as the number of spots per unit area;

[0033] Shape characteristics : By calculating the perimeter of the spot and area The ratio of quantifies the regularity of the spots.

[0034] To further optimize this technical solution, the key characteristic indicators are:

[0035] Color abnormalities include:

[0036] Color contrast :By calculating the Euclidean distance between the abnormal area and the normal area color ;

[0037] Color area ratio : The ratio of the abnormal color area to the total surface area;

[0038] Concave or convex features include:

[0039] Concave and convex depth : Based on the 3D surface reconstruction data, calculate the maximum depth difference of the depression or protrusion;

[0040] Surface roughness : Expressed using RMS height change.

[0041] To further optimize the technical solution, the topological analysis of the fruit skin texture is modeled, and the shape and distribution pattern of the texture anomalies are analyzed based on the defect classification results. The normal fruit skin texture has continuity and distribution characteristics, and the texture in the abnormal area has irregular interruptions, changes or overlaps.

[0042] A real-time online detection system for fruit appearance defects is constructed based on the above-mentioned real-time online detection method for fruit appearance defects. The system comprises an image capture module, a feature extraction and analysis module, a 3D reconstruction module, a dynamic segmentation module, a defect quantification and classification module, and a topology analysis and detection module.

[0043] Further optimizing the technical solution, in the system:

[0044] Image capture module, used to capture multispectral images of the fruit surface to obtain comprehensive and clear appearance information;

[0045] Feature extraction and analysis module, used to realize automatic extraction and cluster analysis of multi-dimensional features of fruit surface;

[0046] 3D reconstruction module, used to reconstruct the 3D model of the fruit surface and enhance the ability to analyze defects in detail;

[0047] Dynamic segmentation module, used to dynamically segment the fruit surface area for real-time defect analysis;

[0048] Defect quantification and grading module, used to quantitatively evaluate the detected defects and grade them according to the comprehensive scores;

[0049] The topological analysis detection module is used to analyze abnormal changes in surface texture through topological analysis.

[0050] Compared with the prior art, the present invention provides a real-time online detection method and system for fruit appearance defects, which have the following beneficial effects:

[0051] This real-time online detection method and system for fruit appearance defects overcomes the shortcomings of traditional methods in detection accuracy, classification precision and environmental adaptability by combining multiple processes such as multispectral image capture, hierarchical feature extraction, high-dimensional clustering analysis, 3D modeling and defect quantitative evaluation. The system can accurately identify various types of defects on the surface of fruits (such as spots, cracks, color abnormalities, etc.) in real time in complex detection environments and quantify their severity. Compared with traditional technologies, this system significantly improves the accuracy, stability and real-time performance of defect detection, providing an efficient and intelligent solution for automatic sorting and quality control of fruits, while reducing labor costs and optimizing overall production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of a process of a real-time online detection method for fruit appearance defects proposed by the present invention;

[0053] Figure 2 The present invention provides a structural schematic diagram of a real-time online detection system for fruit appearance defects. DETAILED DESCRIPTION

[0054] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Embodiment 1:

[0056] See also Figure 1 , a real-time online detection method for fruit appearance defects, comprising:

[0057] Capture multispectral images of fruit surfaces;

[0058] In this embodiment, a multispectral image of the fruit surface is captured. Compared with a common RGB camera, a multispectral camera is used to capture image information of the fruit surface from visible light to near-infrared bands, distinguish color changes on the fruit surface, and detect texture and invisible defects on the fruit surface, including cracks and mold spots.

[0059] Optimize the light source angle, brightness, and polarization of the multispectral camera to reduce interference caused by surface reflections;

[0060] The contrast of different defect features on the fruit surface is enhanced by combining the spectral data of multispectral images in different bands.

[0061] Extract and analyze features of multispectral images;

[0062] In this embodiment, feature extraction and analysis of a multispectral image include:

[0063] Hierarchical feature extraction of multispectral images based on deep learning models;

[0064] High-dimensional clustering analysis is performed on the extracted features to distinguish normal and abnormal areas on the fruit surface.

[0065] Output the marking information of normal and abnormal areas, including the category of abnormality (such as cracks, spots, discoloration, etc.) and the specific location and severity of each abnormality for subsequent grading and processing.

[0066] Furthermore, the layered feature extraction of the multispectral image includes:

[0067] The first layer is extraction, basic texture feature extraction;

[0068] Capture the most intuitive defect features on the fruit surface, such as spots, scratches, cracks, etc. Use the following technologies:

[0069] Local Binary Pattern (LBP): Analyzes the grayscale change pattern of local pixels on the fruit surface to extract texture features. LBP is sensitive to small-scale texture changes and can effectively detect spots and tiny cracks.

[0070] Histogram of Directed Gradients (HOG): This method extracts the directional characteristics of scratches and cracks by analyzing the distribution of surface gradient directions. This method is particularly effective for linear defects.

[0071] Wavelet transform: Decomposes the image into different frequency domains. The high-frequency components can be used to detect areas with rough surfaces or local mutations, and are suitable for identifying fine scratches and burrs.

[0072] The output result is a high-dimensional texture feature vector for subsequent processing.

[0073] The second layer is extraction, color anomaly and gloss change extraction;

[0074] The following method is used to extract the features of color uniformity and gloss reflection anomaly on the fruit surface:

[0075] Color histogram analysis: Calculate the histogram features of each color channel (RGB or Lab color space) on the fruit surface to detect the mutation area in the color distribution. Color anomalies are often manifested as discontinuous or deviation from the main color distribution.

[0076] Gloss detection: By performing polarization correction on multispectral images or using an improved Gaussian model, surface gloss and material characteristics can be separated and abnormal gloss areas can be detected.

[0077] Color clustering and difference detection: Use K-Means or mean shift algorithms to segment the fruit into different color regions and compare the degree of difference in the color region boundaries. Large color changes are usually a sign of discoloration, mold, and other problems.

[0078] The results of this layer extraction are color distribution and gloss distribution features, which are used for further fusion analysis.

[0079] The third layer is extraction, which extracts and analyzes the overall morphological characteristics and consistency of the fruit;

[0080] Shape descriptor extraction: Using methods such as ellipse fitting and Fourier shape description, the aspect ratio, edge smoothness, surface concavity and other features of the fruit are calculated to detect deformation and shape abnormalities.

[0081] Surface consistency detection: The consistency of surface texture direction is detected by texture optical flow algorithm. Abnormal texture flow may indicate obvious depressions or lesions on the fruit skin.

[0082] Global feature fusion: Combining the previously extracted color and texture features, principal component analysis (PCA) is used to compress high-dimensional data and generate a global feature vector for subsequent classification.

[0083] The result of this layer is the feature vector of the whole fruit, which represents the comprehensive morphological and consistency information.

[0084] Generate a 3D model of the fruit surface;

[0085] In this embodiment, in order to further analyze subtle surface defects, a 3D model of the fruit surface can be generated by multi-view shooting technology. Multispectral images of the 2D surface of the fruit are taken at different angles to ensure that the complete surface of the fruit is covered. In each pair of spectral images of adjacent perspectives, algorithms (such as SIFT or SURF) are used to identify and match the key feature points of the surface. According to the geometric parameters of the multispectral camera, the depth value of each feature point is calculated by triangulation to generate a dense point cloud. The point cloud data generated by multiple perspectives are aligned to form a complete 3D model of the fruit surface. By reconstructing the fruit surface with high-precision point cloud data, defects that are difficult to observe with traditional 2D detection, such as tiny depressions or protrusions, can be found. Then, the key areas are analyzed in detail through local magnification technology to magnify the surface features of high-risk areas, helping the system to more accurately judge abnormal areas.

[0086] Perform real-time defect detection and dynamic area segmentation;

[0087] In this embodiment, the fruit is divided into normal areas and possible abnormal areas in real time. The algorithm needs to adaptively adjust the segmentation strategy according to the shape, size and surface characteristics of the fruit. For example, oval fruits require different segmentation methods than spherical fruits. Through dynamic segmentation, reasonable allocation of resources can be achieved, more computing power can be concentrated on suspected abnormal areas, and real-time detection speed can be accelerated.

[0088] Conduct quantitative evaluation and classification modeling of defect characteristics;

[0089] Design a quantitative evaluation system for detected appearance defects, such as the length and width of cracks, the area and distribution density of spots, the contrast of color anomalies, etc. Combined with these features, build a multidimensional classification model to classify defects into different levels such as slight, severe, and unacceptable, and output an analysis report on the nature and degree of harm of the defects in real time, providing a decision-making basis for automatic sorting equipment.

[0090] In this embodiment, the quantitative evaluation and classification modeling of defect characteristics include:

[0091] Key characteristic indicators of the quantitative evaluation system, including crack defect characteristics, spot defect characteristics, color abnormality characteristics, and concave or convex characteristics;

[0092] Construct a multidimensional classification model, which includes a comprehensive scoring function and scoring classification.

[0093] Furthermore, among the key characteristic indicators:

[0094] Crack defect characteristics include:

[0095] length : The pixel distance of the crack in the image is calculated by edge detection method and converted into actual physical length;

[0096] width : Calculated by the width perpendicular to the main direction of the crack, taking the weighted value of the maximum width and the average width;

[0097] Distribution density : Defined as the ratio of the total crack area to the surface area of ​​the fruit:

[0098]

[0099] in, is the total crack area, that is, the sum of the areas of all crack regions, is the total surface area of ​​the fruit.

[0100] Characteristics of spot defects include:

[0101] area : Directly calculate the total number of pixels in the spot area and convert it into the actual physical area;

[0102] density : defined as the number of spots per unit area;

[0103] Shape characteristics : By calculating the perimeter of the spot and area The ratio of quantifies the regularity of the spots:

[0104]

[0105] The closer the value is to 1, the closer the spot shape is to a circle, and the larger the value is, the more irregular the spot shape is.

[0106] Color abnormalities include:

[0107] Color contrast :By calculating the Euclidean distance between the abnormal area and the normal area color :

[0108]

[0109] in, are the pixel values ​​of the red, green, and blue channels of the abnormal area, respectively; They are the pixel values ​​of the red, green, and blue channels of the normal area, respectively.

[0110] Color area ratio : The ratio of the abnormal color area to the total surface area.

[0111] Concave or convex features include:

[0112] Concave and convex depth : Based on the 3D surface reconstruction data, calculate the maximum depth difference of the depression or protrusion;

[0113] Surface roughness : Expressed using RMS height change, defined as:

[0114]

[0115] in, is the total number of sampling points on the surface, is the height of each point on the surface, is the average height.

[0116] The multidimensional classification model classifies fruit defects into three levels: mild (C1), severe (C2), and unacceptable (C3). The core of the classification is to construct a comprehensive scoring function , which takes into account the weights of all defect features:

[0117]

[0118] Among them, the weight Represents the degree of influence of different defect categories on the comprehensive score, which can be optimized through training data. Different weights reflect the importance of defects such as cracks, spots, color anomalies and deformation.

[0119] In the comprehensive scoring function In the weighting criteria of different defect characteristics, the weighting criteria can be set based on the degree of impact of the defect on the overall quality of the fruit. For each defect characteristic weighting criteria, the role of each characteristic in the scoring and the basis for the weight allocation are explained in detail.

[0120] crack :Cracks usually affect the appearance, structure and freshness of fruits. Minor cracks may have little effect on the edibility of fruits, but severe cracks may cause fruits to rot more easily or suffer further damage. Therefore, the length of the cracks should be carefully considered. and width It has a great impact on the score of fruits.

[0121] Weight setting: The weight of the crack Higher, especially for fruits with larger cracks. Reflects the degree of threat that cracks pose to the overall quality of the fruit. Generally, the larger and deeper the cracks are, the more obvious the impact on the quality, and the weight should be increased.

[0122] spot : The number and distribution density of spots and its size Directly affects the appearance and edibility of the fruit. The spots themselves may be caused by external damage, pests and diseases, or natural changes in the ripening process. More and larger spots may lead to reduced consumer acceptance of the fruit.

[0123] Weight setting: The weight of the spot Depends on the fruit variety and market requirements. Some varieties may be acceptable even with spots, while other varieties need to maintain a clean surface. Generally, larger or dense spots will increase the defect grade, so the weight of spots should be moderate.

[0124] Abnormal shape and color : Abnormal shape It usually refers to the dents, bulges or deformations on the surface of the fruit, which may affect the appearance of the fruit and the consumer's purchasing intention. Variations, often caused by uneven ripening, insufficient light or external environmental factors, can also affect the fruit's attractiveness. Abnormal shapes and colors can reduce the fruit's commercial value.

[0125] Weight setting: These defects (especially color abnormalities) often have a greater impact on market sales, especially in markets with high requirements for appearance. Therefore, the weights of shape and color abnormalities are The value should be higher, especially color abnormalities which have a greater impact on the quality of the fruit and are often given priority.

[0126] Surface roughness and cracks : Surface roughness Indicates the smoothness of the fruit's surface texture, which is often directly related to the fruit's ripeness, appearance integrity and taste. A rough surface may be caused by over-ripening, environmental damage or improper handling. Fruits with a high surface roughness may affect their sensory experience.

[0127] Weight setting: weight of surface roughness It can be moderate because although it affects the appearance and feel of the fruit, its impact on the overall quality of the fruit is relatively minor compared to other defects such as cracks and spots. The weight usually needs to be compared with defects such as cracks or color anomalies and is usually lower.

[0128] Based on comprehensive rating The defect levels of the fruit are classified into:

[0129] C1: Slight defect, ;

[0130] C2: Serious defect, ;

[0131] C3: Unacceptable defects, ;

[0132] Among them, the threshold Adjustments can be made through historical test data to ensure that the classification standards meet actual needs.

[0133] When used, the multidimensional classification model includes:

[0134] Data standardization: Normalize all feature values ​​in the quantitative evaluation system so that their value range is unified to [0, 1]. The formula is as follows:

[0135]

[0136] Feature weight optimization: Optimize weights through supervised learning (such as linear regression or gradient descent) , so that the prediction error of the comprehensive scoring function for the defect level in the training samples is minimized.

[0137] Real-time defect grading: Input each fruit’s characteristics into a comprehensive scoring function , calculate the comprehensive score. The value of directly determines the defect level and outputs an analysis report of the defect category (cracks, spots, color abnormalities, etc.).

[0138] Adaptive adjustment: Dynamically adjust the weights in the classification model through feedback learning mechanism and threshold , to adapt to different fruit varieties or new detection needs.

[0139] This model uses a comprehensive scoring function Unifying the feature evaluation of different defect types not only takes into account the independent influence of a single feature, but also realizes the interaction between features through weight design. At the same time, the formulas for quantifying features such as concave-convex depth and color contrast are specially designed according to the physical characteristics of the fruit surface, which is targeted and practical.

[0140] Conduct topological analysis of fruit skin texture;

[0141] In this embodiment, the fruit skin texture topology analysis is modeled in combination with the defect grading results to analyze the shape and distribution pattern of texture anomalies. The normal fruit skin texture has certain continuity and distribution characteristics, while the texture in the abnormal area often has irregular interruptions, changes or overlaps.

[0142] The specific analysis process includes:

[0143] Skin texture feature extraction;

[0144] Use texture analysis algorithms (such as local binary patterns (LBP)) to extract basic features of epidermal texture.

[0145] Topology modeling;

[0146] A topological map of the texture area is constructed, where the continuity of the texture is represented by the connectivity between nodes, and the interruption or change of the defect area is represented by the disconnection of nodes or the change of connectivity.

[0147] The complexity and regularity of the texture are analyzed by calculating the Betti number (i.e., the number of independent loops in the topological graph).

[0148] Normal texture is a continuous topological structure with a low Betti number and regularly distributed loops.

[0149] Abnormal textures include topological breaks, random loops, and significant changes in Betti numbers.

[0150] Texture anomaly detection;

[0151] Detect isolated areas of connectivity loss and breakage in the topology map, identify defect edges and center points. Analyze repeated or overlapping parts in the texture, quantify by node repetition rate and boundary change rate.

[0152] Combined with defect classification results;

[0153] Introducing comprehensive scoring function As a reference, the topological features of the defect area are quantified and used to assist in grading:

[0154] For example, crack defects are identified by analyzing the length and number of topological fractures. For spot defects, the topological isolated area and the crack length and width characteristics are combined. Spot area and density analysis.

[0155] A mapping relationship between defect severity and topological characteristics is established, and the topological analysis results are used as detection reference information.

[0156] Embodiment 2:

[0157] See also Figure 2 Based on the real-time online detection method for fruit appearance defects described in Example 1, a functional module is constructed. The system includes an image capture module, a feature extraction and analysis module, a 3D reconstruction module, a dynamic segmentation module, a defect quantification and grading module, and a topological analysis and detection module.

[0158] The image capture module is used to capture multispectral images of the fruit surface to obtain comprehensive and clear appearance information.

[0159] Equipped with a multispectral camera, covering visible light and near-infrared bands, it enhances the ability to detect different defects. With an optimized optical system, it eliminates surface reflections and light and shadow interference by adjusting the light source angle, brightness and polarization characteristics. It supports dynamic focus function to ensure that the images of fruits of different shapes on the conveyor belt are always clear. It provides high-quality raw data for subsequent analysis, especially for detecting texture details and hidden defects.

[0160] The feature extraction and analysis module is used to realize the automatic extraction and cluster analysis of multi-dimensional features of fruit surface.

[0161] It can realize automatic extraction and cluster analysis of multi-dimensional features of fruit surface, accurately classify and locate the defect types and positions on the fruit surface, and provide support for quantification and grading.

[0162] 3D reconstruction module, used to reconstruct the 3D model of the fruit surface and enhance the ability to analyze defects in detail.

[0163] Generate a high-precision point cloud model of the fruit surface through multi-view shooting technology. Perform 3D reconstruction of suspected defect areas to analyze the depth of cracks, the shape of depressions, etc. Amplify the details of key areas, highlight small defects, and avoid misjudgment. Help detect subtle defects that are difficult to detect with the naked eye or cannot be accurately reflected by two-dimensional images.

[0164] Dynamic segmentation module, used to dynamically segment the fruit surface area for real-time defect analysis.

[0165] Use edge computing technology to dynamically segment the normal area of ​​the fruit from the suspected defect area. Automatically adjust the segmentation strategy for fruits of different shapes such as elliptical and spherical shapes. Provide real-time defect detection results, including the location, area and type of the defect.

[0166] The defect quantification and grading module is used to quantitatively evaluate the detected defects and grade them according to the comprehensive scores.

[0167] Quantify the length and width of cracks, the area and density of spots, the contrast of color anomalies, etc. Use a comprehensive scoring function to calculate the overall score of defects based on the weights. Output defect classification results (minor, severe, unacceptable) and generate real-time analysis reports.

[0168] The topological analysis detection module is used to detect abnormal changes in surface texture through topological analysis.

[0169] Build a topological model of the fruit surface texture and analyze the continuity and regularity of the texture. Automatically identify irregular texture changes, such as the extension of surface cracks or the diffusion area of ​​spots. Provide detailed information on texture anomalies, including distribution pattern and severity. Further improve the detection accuracy and stability of complex surface anomalies.

[0170] After the above modules, the detection results are finally output.

[0171] The beneficial effects of the present invention are:

[0172] This real-time online detection system for fruit appearance defects overcomes the shortcomings of traditional methods in detection accuracy, classification precision and environmental adaptability by combining multiple processes such as multispectral image capture, hierarchical feature extraction, high-dimensional clustering analysis, 3D modeling and defect quantitative evaluation. The system can accurately identify various types of defects on the surface of fruits (such as spots, cracks, color abnormalities, etc.) in real time in complex detection environments and quantify their severity. Compared with traditional technologies, this system significantly improves the accuracy, stability and real-time performance of defect detection, providing an efficient and intelligent solution for automatic sorting and quality control of fruits, while reducing labor costs and optimizing overall production efficiency.

[0173] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0174] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time online detection method for fruit appearance defects, characterized in that: include: Capture multispectral images of fruit surfaces; Extract and analyze features of multispectral images to distinguish normal and abnormal areas on the fruit surface; Generate a 3D model of the fruit surface based on multispectral images and zoom in on abnormal areas in detail; Perform real-time defect detection and dynamic area segmentation, and perform defect feature detection and area segmentation on the fruit surface; Conduct quantitative evaluation and classification modeling of defect characteristics and output defect classification results; Conduct topological analysis of fruit skin texture and analyze abnormal textures in combination with defect grading results.

2. A real-time online detection method for fruit appearance defects according to claim 1, characterized in that: Capture multispectral images of the fruit surface, including: Use a multispectral camera to capture image information from visible light to near-infrared bands on the fruit surface, identify color changes on the fruit surface, and detect texture and invisible defects on the fruit surface, including cracks and mold spots; Optimize the light source angle, brightness, and polarization of the multispectral camera to reduce interference caused by surface reflections; The contrast of different defect features on the fruit surface is enhanced by combining the spectral data of multispectral images in different bands.

3. A real-time online detection method for fruit appearance defects according to claim 1, characterized in that: When extracting and analyzing features of multispectral images, it includes: Hierarchical feature extraction of multispectral images based on deep learning models; High-dimensional clustering analysis is performed on the extracted features to distinguish normal and abnormal areas on the fruit surface.

4. A real-time online detection method for fruit appearance defects according to claim 3, characterized in that: When performing hierarchical feature extraction on multispectral images, it includes: The first layer is extraction, basic texture feature extraction; The second layer is extraction, color anomaly and gloss change extraction; The third layer is extraction, which extracts and analyzes the overall morphological characteristics and consistency of the fruit.

5. A real-time online detection method for fruit appearance defects according to claim 1, characterized in that: When quantitatively evaluating and classifying defect characteristics, the following aspects should be considered: Key characteristic indicators of the quantitative evaluation system, including crack defect characteristics, spot defect characteristics, color abnormality characteristics, and concave or convex characteristics; Construct a multidimensional classification model, which includes a comprehensive scoring function and scoring classification.

6. A real-time online detection method for fruit appearance defects according to claim 5, characterized in that: Among the key characteristic indicators: Crack defect characteristics include: length : The pixel distance of the crack in the image is calculated by edge detection method and converted into actual physical length; width : Calculated by the width perpendicular to the main direction of the crack, taking the weighted value of the maximum width and the average width; Distribution density : defined as the ratio of the total crack area to the fruit surface area; Characteristics of spot defects include: area : Directly calculate the total number of pixels in the spot area and convert it into the actual physical area; density : defined as the number of spots per unit area; Shape characteristics : By calculating the perimeter of the spot and area The ratio of quantifies the regularity of the spots.

7. A real-time online detection method for fruit appearance defects according to claim 5, characterized in that: Among the key characteristic indicators: Color abnormalities include: Color contrast :By calculating the Euclidean distance between the abnormal area and the normal area color ; Color area ratio : The ratio of the abnormal color area to the total surface area; Concave or convex features include: Concave and convex depth : Based on the 3D surface reconstruction data, calculate the maximum depth difference of the depression or protrusion; Surface roughness : Expressed using RMS height change.

8. A real-time online detection method for fruit appearance defects according to claim 1, characterized in that: The fruit skin texture topology analysis is modeled, and the shape and distribution pattern of texture anomalies are analyzed based on the defect classification results. The normal fruit skin texture has continuity and distribution characteristics, and the texture in the abnormal area has irregular interruptions, changes or overlaps.

9. A real-time online detection system for fruit appearance defects, constructed based on the real-time online detection method for fruit appearance defects according to any one of claims 1 to 8, characterized in that: The system includes an image capture module, a feature extraction and analysis module, a 3D reconstruction module, a dynamic segmentation module, a defect quantification and grading module, and a topology analysis and detection module.

10. A real-time online detection system for fruit appearance defects according to claim 9, characterized in that: The system comprises: Image capture module, used to capture multispectral images of the fruit surface to obtain comprehensive and clear appearance information; Feature extraction and analysis module, used to realize automatic extraction and cluster analysis of multi-dimensional features of fruit surface; 3D reconstruction module, used to reconstruct the 3D model of the fruit surface and enhance the ability to analyze defects in detail; Dynamic segmentation module, used to dynamically segment the fruit surface area for real-time defect analysis; Defect quantification and grading module, used to quantitatively evaluate the detected defects and grade them according to the comprehensive scores; The topological analysis detection module is used to analyze abnormal changes in surface texture through topological analysis.

Images