X-ray-based walnut internal defect feature optimization detection method and device

Abstract

The invention relates to the field of nondestructive testing and automatic sorting of agricultural products, and discloses an X-ray-based walnut internal defect feature optimization detection method and device, and the method comprises the steps: collecting an X-ray image of a moving walnut, extracting a region of interest (ROI) through preprocessing, extracting and fusing spatial domain and frequency domain texture features to form a multi-dimensional feature set, and obtaining a feature set; an optimal feature combination is screened out through dimensionality reduction; and finally, the types of the internal defects of the walnuts are identified by a pre-trained classifier, and a result is output. The device comprises a feeding mechanism, a conveying mechanism, an X-ray detection mechanism, a sorting execution mechanism and a processing and control system electrically connected with all the mechanisms. The problems of single identification category and insufficient model stability and precision are solved, and high-throughput and automatic lossless sorting of walnuts is realized.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing and automated sorting technology for agricultural products, specifically to a method and apparatus for the optimal detection of internal defect features in walnuts based on X-rays. Background Technology

[0002] Walnut quality sorting is a crucial step in the walnut processing process. In particular, the ability to accurately identify and sort good, empty, and moldy walnuts in one go is related to the efficiency and safety of the entire processing process. During the harvesting, hulling, and transportation of walnuts, the shells are easily damaged, leading to internal defects such as emptiness or mold. If moldy walnuts are not identified in time during circulation, it will not only affect product quality but also pose a potential threat to consumers' health.

[0003] Currently, walnut quality sorting relies heavily on manual screening. Traditional testing methods are often destructive, requiring the shell to be opened and the kernel removed for internal quality assessment, rendering the sample unusable. Such methods are not only time-consuming and labor-intensive but also costly, failing to meet the urgent need for automated and non-destructive testing technologies in the current large-scale, high-efficiency processing conditions of the walnut industry.

[0004] In addition, although there are existing methods for detecting walnut damage based on acoustics, assessing external quality based on machine vision, and identifying internal defects by combining X-ray images and deep learning, these technologies generally suffer from problems such as a single detection target and insufficient attention to internal quality. Moreover, most methods can only identify one type of defect, making it difficult to achieve efficient sorting of good, empty, and moldy fruits across multiple categories. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an X-ray-based method and apparatus for the optimal detection of internal defects in walnuts. The aim is to solve the problems of existing technologies in detecting internal defects in walnuts, such as the limited identification of categories, the inability to effectively distinguish between multiple defect types such as shriveled or moldy walnuts, and the insufficient accuracy and stability of classification models.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides a method for optimal detection of internal defect features in walnuts based on X-rays, comprising the following steps:

[0008] S1. Acquire X-ray images of walnuts in motion;

[0009] S2. Preprocess the X-ray image and extract the region of interest (ROI) containing the walnut body;

[0010] S3. Extract spatial domain texture features and frequency domain texture features from the region of interest (ROI), and fuse the features into a multi-dimensional fused feature set;

[0011] S4. Perform dimensionality reduction processing on the multidimensional fusion feature set to select the optimal feature combination;

[0012] S5. Based on the optimal feature combination, a pre-trained classifier is used to identify the types of internal defects in walnuts and output the classification results.

[0013] Preferably, the preprocessing in step S2 specifically includes:

[0014] The acquired X-ray images were cropped in the center and normalized in grayscale.

[0015] Preferably, the process of extracting the region of interest (ROI) in step S2 includes:

[0016] The Otsu adaptive threshold segmentation method was used to extract the foreground region of the walnut, morphological operations were used to eliminate noise, and the edge correction of the segmented foreground was performed by the convex hull algorithm.

[0017] Preferably, the texture features extracted in step S3 include spatial domain texture features and frequency domain texture features.

[0018] Preferably, the spatial domain texture features include basic texture statistical features extracted based on gray-level histograms and texture features extracted based on gray-level co-occurrence matrix (GLCM), gray-level difference matrix (GLDM), and local binary pattern (LBP); the frequency domain texture features are extracted based on Fourier transform (FT) and wavelet transform (WT).

[0019] The frequency domain texture features are extracted based on Fourier transform (FT) and wavelet transform (WT).

[0020] Preferably, the feature selection algorithm used in step S4 is at least one of the Progressive Variable Elimination (IRIV) algorithm or the Variable Importance Index (VISSA) algorithm.

[0021] Preferably, the classification model used in step S5 is at least one of logistic regression (LR), k-nearest neighbors (KNN), or extreme learning machine (ELM).

[0022] Preferably, step S5 further includes:

[0023] Before training the classification model, the hyperparameters of the classification model are optimized using the Grey Wolf Optimization Algorithm (GWO), Particle Swarm Optimization Algorithm (PSO), or Arithmetic Optimization Algorithm (AOA).

[0024] Preferably, step S4 employs a strategy combining the Variable Importance Index Screening (VISSA) algorithm with the Incremental Variable Elimination (IRIV) method for dimensionality reduction.

[0025] Step S5 uses a logistic regression (LR) model optimized by the Grey Wolf (GWO) algorithm for identification.

[0026] A second aspect of the present invention provides an X-ray-based preferred detection device for internal defect features of walnuts, the device comprising:

[0027] The feeding mechanism is used to place the walnuts to be tested onto the conveyor mechanism;

[0028] The conveying mechanism includes a conveyor belt driven by a DC motor;

[0029] An X-ray detection mechanism, housed in a lead box, includes an X-ray emission source and a flat panel detector.

[0030] The first photoelectric sensor and the second photoelectric sensor are set on the path of the conveyor belt to detect the position of the walnuts;

[0031] The sorting actuator includes a solenoid valve control unit and a push rod cylinder pneumatically connected to an air compressor.

[0032] An abnormal fruit collection trough and a good fruit collection container are located downstream of the sorting actuator;

[0033] The processing and control system includes an industrial computer and a PLC, and the processing and control system is electrically connected to the X-ray detection mechanism, the first photoelectric sensor, the second photoelectric sensor and the sorting execution mechanism;

[0034] The first photoelectric sensor is used to trigger the X-ray detection mechanism to acquire images when a walnut is detected. The processing and control system generates sorting instructions based on the acquired images. The second photoelectric sensor is used to trigger the PLC to control the push rod cylinder to move according to the sorting instructions when an detected walnut is detected, so as to push the walnut into the corresponding abnormal fruit collection slot or good fruit collection container.

[0035] This invention provides a method and apparatus for optimal detection of internal defects in walnuts based on X-rays.

[0036] It has the following beneficial effects:

[0037] 1. This invention integrates spatial domain texture features that describe the spatial distribution relationship of pixels with frequency domain texture features that reveal the periodicity of textures, constructing a multi-dimensional, complementary feature set. Compared with existing technologies that rely on a single feature dimension for analysis, this invention provides an effective way to more fully characterize the subtle differences in the internal structure of walnuts, helping to improve the recognition accuracy and generalization ability of various internal states such as good fruit, empty fruit, and moldy fruit.

[0038] 2. This invention introduces a two-stage feature selection strategy combining VISSA and IRIV. Through progressive dimensionality reduction, a more representative subset of core features is selected. This is of positive significance for improving the model overfitting and performance instability problems that may be caused by direct modeling due to feature redundancy. It helps to build a more concise and efficient classification model while ensuring classification accuracy.

[0039] 3. In the classifier construction stage, this invention uses intelligent algorithms such as Grey Wolf Optimization (GWO) to systematically optimize the hyperparameters of the classification model. This approach makes it possible to get rid of the dependence on human experience, explore and approach the theoretical upper limit of classifier performance, and help to further explore and utilize the inherent potential of the selected classification algorithm in order to obtain better defect recognition results.

[0040] 4. This invention organically combines efficient X-ray non-destructive imaging technology with optimized machine learning algorithms to form a closed-loop automated detection process from online image acquisition to final intelligent classification. Compared with traditional destructive manual detection methods, it provides a high-throughput, non-destructive automated processing solution, offering a useful attempt and reference for solving problems such as low efficiency, strong subjectivity of results, and difficulty in adapting to the pace of modern production in manual detection. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0042] Figure 2 This is a schematic diagram of the working process of the device of the present invention;

[0043] Figure 3 A flowchart is established for the model of this invention;

[0044] Figure 4 This is a schematic diagram of the preferred detection device for internal defects in walnuts based on X-rays according to the present invention.

[0045] The components include: 1. Good fruit collection container; 2. Push rod cylinder; 3. Conveyor belt; 4. Second photoelectric sensor; 5. X-ray emission source; 6. Lead box; 7. Flat panel detector; 8. First photoelectric sensor; 9. DC motor; 10. Feeding mechanism; 11. Industrial control computer; 12. PLC; 13. Air compressor; 14. Solenoid valve control unit; and 15. Abnormal fruit collection trough. Detailed Implementation

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see the appendix Figure 1 - Appendix Figure 3 This invention provides an X-ray-based method for optimal detection of internal defects in walnuts, comprising the following steps:

[0048] S1. Acquire X-ray images of walnuts in motion;

[0049] Specifically, this step first involves introducing the walnuts to be inspected one by one into a conveyor path at a preset interval, and making them move continuously along the path at a basically constant speed. This is to ensure that each walnut can be imaged independently and clearly, avoiding overlap or missed detection.

[0050] Subsequently, the moving walnut is guided through a pre-defined X-ray detection area, which is spatially defined as the effective range between the X-ray emitting end and the image receiving end.

[0051] By detecting the movement of the walnut online, a synchronous trigger signal is generated instantly when it reaches the precise trigger point within the detection area. This signal is used to initiate a complete imaging operation.

[0052] Based on the received trigger signal, the X-ray source is excited in a very short time, causing it to emit a beam of X-rays with specific energy and dose. The direction of the beam is precisely controlled to ensure that it can completely penetrate the walnut body that has moved to the trigger point.

[0053] After X-rays penetrate a walnut, their intensity will decrease depending on the density and material of the walnut's interior, such as the fullness of the kernel, the presence of hollows or mold, etc. An image sensing device on the other side receives this attenuated X-ray beam carrying information about the walnut's internal structure.

[0054] Finally, the image sensing device converts the received X-ray intensity signal into a two-dimensional digital electrical signal and further quantizes it to generate and acquire a frame of digital X-ray image. In this image, different gray values ​​directly correspond to the density differences at different locations inside the walnut.

[0055] This embodiment achieves precise positioning and instantaneous X-ray imaging of a single walnut in motion during an automated conveying process using the method described above. This results in a digital image that clearly reflects the internal density differences, laying a data foundation for subsequent defect feature analysis.

[0056] S2. Preprocess the X-ray image and extract the region of interest (ROI) containing the walnut body;

[0057] Specifically, this step aims to remove image background and noise, and accurately segment the walnut region.

[0058] First, the central region of the acquired image is cropped to remove background edges that are unrelated to the walnut subject.

[0059] Subsequently, grayscale normalization is performed, linearly mapping pixel values ​​to a uniform grayscale range, such as [0, 255], to eliminate differences caused by fluctuations in imaging conditions. The calculation formula is as follows:

[0060]

[0061] In the formula, I norm (x,y) represents the pixel gray value at coordinates (x,y) in the normalized image; I(x,y) represents the pixel gray value at coordinates (x,y) in the original image; I max I represents the maximum pixel grayscale value in the original image. min This is the minimum pixel grayscale value in the original image.

[0062] After preprocessing, the Otsu adaptive thresholding method is used to automatically calculate the optimal threshold based on the grayscale histogram of the image, thus separating the foreground and background of the walnut subject.

[0063] To optimize the segmentation results, morphological operations such as opening and closing operations are further used to optimize the binarized image, in order to eliminate isolated noise and fill in tiny holes inside the foreground region.

[0064] Finally, to obtain a smooth and complete walnut outline, the convex hull algorithm is used to correct the edges of the segmented foreground, ultimately obtaining an accurate region of interest (ROI) that contains only the main body of the walnut.

[0065] S3. Extract spatial domain texture features and frequency domain texture features from the region of interest (ROI), and fuse the features into a multi-dimensional fused feature set;

[0066] Specifically, in order to comprehensively describe the internal texture information of walnuts, this step extracts two types of features in parallel: spatial domain and frequency domain, from the region of interest (ROI).

[0067] Spatial domain features describe the spatial distribution relationship of pixel gray levels, including features extracted based on the gray-level co-occurrence matrix (GLCM), the gray-level difference matrix (GLDM), and the local binary pattern (LBP). The definition of GLCM is as follows:

[0068] P(i,j,d,θ)=#{(x1,y1),(x2,y2)∈S|I(x1,y1)=i;

[0069] In the formula, P(i,j,d,θ) is the value of the gray-level co-occurrence matrix (GLCM); i,j are the specified pixel gray levels; d is the spatial distance between pixel pairs; θ is the directional angle of the pixel pair; # is a counting symbol, indicating the number of pixel pairs that meet specific conditions; (x1,y1) and x2,y2 are the coordinates of a pixel pair; S is the set of all pixel coordinates in the image or region of interest (ROI); I(x1,y1) is the gray value of the pixel at coordinates x1,y1.

[0070] GLDM performs statistical analysis based on the gray-level differences between pixel pairs, and its difference probability density function is defined as:

[0071]

[0072] In the formula, P d (k) represents the probability that the gray level difference is kk; k is the gray level difference of the pixel; I(x,y) is the gray level value of the pixel at coordinates (x,y); Δx,y+Δy is the spatial displacement determined by distance and direction; # is a counting symbol, indicating the number of pixels that meet the conditions; N p This represents the total number of pixel pairs in the image.

[0073] LBP describes local texture by comparing the gray values ​​of the center pixel with those of its neighboring pixels. The calculation formula is as follows:

[0074]

[0075] In the formula, LBP P,R For the center pixel (x) c ,y c The LBP value calculated at () is given; P is the number of pixels in the neighborhood; R is the radius of the neighborhood; g c The grayscale value of the center pixel; g p Let be the gray value of the p-th neighboring pixel; s(z) is a threshold function.

[0076] Frequency domain texture features are obtained by performing Fourier transform (FT) and wavelet transform (WT) on the ROI to extract features such as energy and entropy from different frequencies and scales in order to capture the periodic information of the texture.

[0077] Finally, all extracted spatial and frequency domain features are concatenated and fused into a multidimensional fusion feature set that can comprehensively characterize the internal state of the walnut.

[0078] S4. Perform dimensionality reduction on the multidimensional fusion feature set to select the optimal feature combination;

[0079] Specifically, the multidimensional fusion feature set constructed in S3 has a high dimensionality and may contain redundant or irrelevant features. Directly using it for modeling will affect efficiency and accuracy, so dimensionality reduction is necessary.

[0080] This step uses a feature selection algorithm to achieve this purpose. Preferably, at least one of the progressive variable elimination method (IRIV) or the variable importance index selection algorithm (VISSA) can be used.

[0081] In a more preferred embodiment, a strategy combining VISSA and IRIV is adopted. First, the VISSA algorithm is used for coarse screening to quickly eliminate a large number of irrelevant features.

[0082] Then, the IRIV algorithm is used to refine the remaining features, and the variables with the least contribution are eliminated through iteration, and finally the optimal feature combination containing the strongest classification information is selected.

[0083] S5. Based on the optimal feature combination, a pre-trained classifier is used to identify the types of internal defects in walnuts and output the classification results.

[0084] Specifically, this step is the decision-making stage of the method. First, it is necessary to prepare a labeled sample dataset, such as one that has been distinguished into good fruit, empty fruit, and moldy fruit, and then perform S1 to S4 on it to extract the optimal features.

[0085] Subsequently, a classification model is selected for training, which can be at least one of the following: logistic regression (LR), k-nearest neighbors (KNN), or extreme learning machine (ELM).

[0086] To improve the model's classification performance, optimization algorithms are preferred to optimize the model's key hyperparameters before formal training. Optimization algorithms include Grey Wolf Optimization (GWO), Particle Swarm Optimization (PSO), or Arithmetic Optimization Algorithm (AOA).

[0087] The logistic regression (LR) model optimized by the Grey Wolf (GWO) algorithm is used as the final classifier.

[0088] By inputting the optimal feature combination of the walnut to be tested into this pre-trained model, a classification result representing its internal defect type can be output, such as "good fruit", "empty fruit", or "moldy fruit".

[0089] Please see the appendix Figure 4 This invention provides an X-ray-based device for the optimal detection of internal defects in walnuts, comprising:

[0090] Feeding mechanism 10 is used to place the walnuts to be tested onto the conveying mechanism;

[0091] The conveying mechanism includes a conveyor belt 3 driven by a DC motor 9;

[0092] The X-ray detection mechanism is housed in a lead box 6. The X-ray detection mechanism includes an X-ray emission source 5 and a flat panel detector 7.

[0093] The first photoelectric sensor 8 and the second photoelectric sensor 4 are set on the path of the conveyor belt 3 to detect the position of the walnuts;

[0094] The sorting actuator includes a solenoid valve control unit 14 and a push rod cylinder 2 that is pneumatically connected to an air compressor 13;

[0095] Abnormal fruit collection tank 15 and good fruit collection container 1 are located downstream of the sorting actuator;

[0096] The processing and control system includes an industrial computer 11 and a PLC 12. The processing and control system is electrically connected to the X-ray detection mechanism, the first photoelectric sensor 8, the second photoelectric sensor 4, and the sorting actuator.

[0097] The first photoelectric sensor 8 is used to trigger the X-ray detection mechanism to acquire images when a walnut is detected. The processing and control system generates sorting instructions based on the acquired images. The second photoelectric sensor 4 is used to trigger the PLC 12 to control the push rod cylinder 2 to move according to the sorting instructions when a detected walnut is detected, so as to push the walnut into the corresponding abnormal fruit collection slot 15 or good fruit collection container 1.

[0098] Working principle: The working process of this device begins with the feeding mechanism 10, which places the walnuts to be tested in an orderly and individual manner on the conveyor belt 3 of the conveyor mechanism driven by the DC motor 9. The walnuts enter the detection area at a constant speed with the conveyor belt 3. When they pass the first photoelectric sensor 8 on the path, the sensor immediately generates a trigger signal. This signal notifies the processing and control system on the one hand, and activates the X-ray detection mechanism contained in the lead box 6 on the other hand, so that it enters the standby state.

[0099] Subsequently, when the walnut moves precisely to the imaging window, the X-ray emission source 5 and the flat panel detector 7 are activated synchronously to complete the penetrating imaging of the walnut. The acquired X-ray image is transmitted in real time to the industrial control computer 11 in the processing and control system. The industrial control computer 11 calls the preset detection method to analyze and process the image, completes all steps from feature extraction, selection to final classification, and generates sorting instructions that characterize the internal state of the walnut (such as good fruit or abnormal fruit).

[0100] The sorting command is sent from the industrial control computer 11 to the PLC 12 for temporary storage. When the detected walnut continues to move along the conveyor belt 3 and is detected by the second photoelectric sensor 4 at the end of the path, the PLC 12 is triggered. The PLC 12 immediately controls the solenoid valve control unit 14 in the sorting execution mechanism according to the received sorting command, and then drives the push rod cylinder 2, which is pneumatically connected to the air compressor 13, to perform precise actions.

[0101] Finally, the pusher cylinder 2 pushes the walnuts into the designated collection container. Good walnuts fall into the good walnut collection container 1, while abnormal walnuts are guided into the abnormal walnut collection trough 15. This realizes fully automatic, high-precision online non-destructive detection and sorting of internal defects in walnuts.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimal detection of internal defects in walnuts based on X-rays, characterized in that, Includes the following steps: S1. Acquire X-ray images of walnuts in motion; S2. Preprocess the X-ray image and extract the region of interest (ROI) containing the walnut body; S3. Extract spatial domain texture features and frequency domain texture features from the region of interest (ROI), and fuse the features into a multi-dimensional fused feature set; S4. Perform dimensionality reduction processing on the multidimensional fusion feature set to select the optimal feature combination; S5. Based on the optimal feature combination, a pre-trained classifier is used to identify the types of internal defects in walnuts and output the classification results.

2. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, The preprocessing in step S2 specifically includes: The acquired X-ray images were cropped in the center and normalized in grayscale.

3. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, The process of extracting the region of interest (ROI) in step S2 includes: The Otsu adaptive threshold segmentation method was used to extract the foreground region of the walnut, morphological operations were used to eliminate noise, and the edge correction of the segmented foreground was performed by the convex hull algorithm.

4. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, The texture features extracted in step S3 include spatial domain texture features and frequency domain texture features.

5. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, The spatial domain texture features include basic texture statistical features extracted based on gray-level histograms, as well as texture features extracted based on gray-level co-occurrence matrix (GLCM), gray-level difference matrix (GLDM), and local binary pattern (LBP); the frequency domain texture features are extracted based on Fourier transform (FT) and wavelet transform (WT).

6. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, The feature selection algorithm used in step S4 is at least one of the Progressive Variable Elimination Method (IRIV) or the Variable Importance Index Selection Algorithm (VISSA).

7. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, The classification model used in step S5 is at least one of logistic regression (LR), k-nearest neighbors (KNN), or extreme learning machine (ELM).

8. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, Step S5 further includes: Before training the classification model, the hyperparameters of the classification model are optimized using the Grey Wolf Optimization Algorithm (GWO), Particle Swarm Optimization Algorithm (PSO), or Arithmetic Optimization Algorithm (AOA).

9. The method for optimal detection of internal defects in walnuts based on X-rays according to claim 1, characterized in that, Step S4 employs a strategy combining the Variable Importance Index Screening (VISSA) algorithm with the Progressive Variable Elimination (IRIV) method for dimensionality reduction. Step S5 uses a logistic regression (LR) model optimized by the Grey Wolf (GWO) algorithm for identification.

10. An X-ray-based device for optimizing the detection of internal defects in walnuts, and a method for optimizing the detection of internal defects in walnuts based on X-rays according to any one of claims 1-9, characterized in that, The device includes: Feeding mechanism (10) is used to place the walnuts to be tested onto the conveying mechanism; The conveying mechanism includes a conveyor belt (3) driven by a DC motor (9); An X-ray detection mechanism is housed in a lead box (6), the X-ray detection mechanism including an X-ray emission source (5) and a flat panel detector (7); The first photoelectric sensor (8) and the second photoelectric sensor (4) are set on the path of the conveyor belt (3) to detect the position of the walnut; The sorting actuator includes a solenoid valve control unit (14) and a push rod cylinder (2) pneumatically connected to an air compressor (13); An abnormal fruit collection trough (15) and a good fruit collection container (1) are disposed downstream of the sorting execution mechanism; The processing and control system includes an industrial computer (11) and a PLC (12), and the processing and control system is electrically connected to the X-ray detection mechanism, the first photoelectric sensor (8), the second photoelectric sensor (4) and the sorting execution mechanism; The first photoelectric sensor (8) is used to trigger the X-ray detection mechanism to acquire images when a walnut is detected. The processing and control system generates sorting instructions based on the acquired images. The second photoelectric sensor (4) is used to trigger the PLC (12) to control the push rod cylinder (2) to move according to the sorting instructions when a detected walnut is detected, so as to push the walnut into the corresponding abnormal fruit collection trough (15) or good fruit collection container (1).

Images