A Non-destructive Testing Device and Method for Nut Quality Based on Fully Automated Miniature CT

Abstract

This invention provides a non-destructive testing device and method for nut quality based on a fully automated micro-CT scanner. The method includes: S1, using a transport mechanism to transport the nuts to a rotating platform; S2, keeping the micro-focal spot X-ray source and the flat panel detector stationary, and controlling the rotating platform to rotate intermittently via a PLC controller to achieve full-angle irradiation of the nuts by the micro-focal spot X-ray source; S3, during the intermittent stops of the rotating platform, acquiring planar array images of the nuts from various angles using the flat panel detector to obtain a full-angle planar array image of the nuts, and transmitting it to a computer; S4, the computer performs tomographic reconstruction and image processing based on the full-angle planar array image to obtain the quality trait parameters of the nuts. This invention enables non-destructive testing of nut quality trait parameters through visual recognition technology, significantly improving testing accuracy and efficiency compared to manual trait testing methods and traditional culture methods for detecting mold infection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field, in particular to a kind of based on automatic micro-CT's nut quality nondestructive testing device and detection method. BACKGROUND

[0002] With the rapid development and extensive use of plant fruit character detection technology, people use fruit character detection technology to detect the character of fruit gradually becomes the mainstream of fruit character detection technology application. But in the process of fruit character detection, the existing fruit character detection method relies on manual calculation of fruit flesh quality, manual peeling of fruit skin, using vernier caliper to measure the thickness of fruit skin, and taking the average value of the thickest and thinnest as the thickness of fruit skin. The precision and efficiency of this detection method are low, and people hope to improve the precision and efficiency of fruit character detection. Therefore, how to intelligently detect the character of fruit to improve the precision and efficiency of fruit character detection is the goal pursued. Nut food is accepted by more and more consumers due to its comprehensive and rich nutritional characteristics. With the continuous improvement of current economic level and the continuous attention to food safety, the quality of nut food gradually becomes the basis for consumers to choose brands. However, in recent years, the phenomenon of mold exceeding standard in nut food is common in market supervision sampling, which has a great impact on human health and social economy. At present, the detection of nut quality and mold infection is still the traditional culture method, which gradually becomes the difficulty and pain point of industry development due to the problems of complicated operation steps and poor timeliness of detection. SUMMARY

[0003] The purpose of the present application is to overcome the defects of the prior art, and to provide a kind of based on automatic micro-CT's nut quality nondestructive testing device and detection method, which can detect the quality character parameters of nuts by visual recognition, and the detection precision and efficiency are improved compared with manual detection.

[0004] To achieve the above-mentioned purpose, the present application adopts the following specific technical scheme:

[0005] The application provides a nut quality nondestructive testing device based on a full-automatic micro-CT, which comprises a load rotating table, a carrying mechanism, a flat panel detector, a micro-focus ray source and a lifting table.

[0006] Preferably, the computer comprises a device control module, an image acquisition module, an image processing module and a quality characteristic calculation unit.

[0007] Preferably, the image processing module comprises a tomographic image reconstruction unit, a semantic segmentation unit, a three-dimensional reconstruction unit and the quality characteristic calculation unit.

[0008] Preferably, the image processing module further comprises an image preprocessing unit.

[0009] Preferably, the carrying mechanism is a mechanical hand.

[0010] The application provides a nut quality nondestructive detection method based on a full-automatic micro-CT.

[0011] S1, using a carrying mechanism to carry nuts to a load rotating table;

[0012] S2, keeping a micro-focus spot ray source and a flat panel detector static, controlling the load rotating table to rotate intermittently through a PLC controller, and realizing full-angle irradiation of the micro-focus spot ray source on the nuts;

[0013] S3, when the load rotating table is intermittently stopped, collecting a surface array image of each angle of the nuts through the flat panel detector, obtaining a full-angle surface array image of the nuts, and transmitting the full-angle surface array image to a computer;

[0014] S4, performing tomographic reconstruction and image processing on the full-angle surface array image based on the computer, and obtaining quality trait parameters of the nuts.

[0015] Preferably, in step S4, the process of performing tomographic reconstruction on the full-angle surface array image comprises the following steps:

[0016] S41, extracting a sinogram composed of the same row from each image of the full-angle surface array image;

[0017] S42, performing reconstruction of the tomographic image by using a three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm based on a CUDA architecture, and performing noise reduction processing on the tomographic image in combination with ray offset calibration and gain calibration; wherein,

[0018] The formula of the three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm based on the CUDA architecture is as follows:

[0019]

[0020] Wherein, f(x, y, z) is a density function of the nuts, β is an included angle between a normal direction of the flat panel detector and a y-axis of a nut coordinate system, and R is a vertical distance between a coordinate origin of the nut coordinate system and a connecting line between the micro-focus spot ray source and a projection position of the reconstructed point on the flat panel detector.

[0021] Preferably, in step S4, the process of performing image processing on the tomographic image comprises the following steps:

[0022] S43, selecting part of the tomographic images as a training set, generating different types of labels by using an image label generation method, and marking a shell region, a pulp region and a mildew region in the data set;

[0023] S44, select the tomographic image with the label having a matching degree greater than the threshold value as a training set to train the deep learning semantic segmentation model based on the convolutional neural network;

[0024] S45, perform semantic segmentation on all tomographic images by the deep learning semantic segmentation model to obtain a semantic segmentation result;

[0025] S46, based on the semantic segmentation result, reconstruct a three-dimensional model of the nut to obtain a three-dimensional digital model of the nut;

[0026] S47, based on the three-dimensional digital model of the nut, calculate the quality trait parameters of the nut.

[0027] Preferably, step S43 specifically comprises the following steps:

[0028] S431, analyze different tissue features of the nut from the perspectives of gray scale, gradient and local contrast;

[0029] S432, design a feature extraction operator according to the analysis result to distinguish different tissues of the nut;

[0030] S433, set a class label to complete the labeling of the shell region, the pulp region and the moldy region.

[0031] Preferably, in step S46, the three-dimensional digital model of the nut is obtained by equal-step sequential stacking using the semantic segmentation result of the tomographic image.

[0032] Preferably, in step S47, the quality trait parameters of the nut include nut size, nut surface area, nut volume, nut morphological characteristics, pulp size, pulp volume, shell thickness, pulp ratio and moldy volume.

[0033] The calculation method of the nut size is as follows:

[0034] The length and width of the nut are obtained by fitting an excircle ellipse to the three-dimensional digital model to solve the major axis and minor axis of the smallest fitting ellipse.

[0035] The calculation method of the pulp size is as follows:

[0036] The length and width of the pulp are obtained by fitting an excircle ellipse to the pulp middle layer of the three-dimensional digital model to solve the major axis and minor axis of the smallest fitting ellipse.

[0037] The calculation method of the nut surface area is as follows:

[0038] The surface area of the nut is calculated by analyzing the connected domain of the outermost voxels of the three-dimensional digital model.

[0039] The calculation methods of the nut volume, pulp volume and moldy volume are as follows:

[0040] The volume of the fruit, the flesh and the mold part is obtained by multiplying the voxel set of different tissues of the nut with the actual size corresponding to each tissue voxel.

[0041] The calculation method of the proportion of the flesh is as follows:

[0042] The proportion of the flesh in the volume of the nut is obtained by dividing the volume of the flesh by the volume of the nut;

[0043] The calculation method of the shell thickness is as follows:

[0044] The thickness of the shell is obtained by solving the shortest distance between the inner and outer contours of the shell;

[0045] The judgment method of the morphological characteristics of the nut is as follows:

[0046] The closer the ratio of the length and the width of the nut is to one, the closer the nut is to a circle.

[0047] Preferably, before step S43, an image preprocessing step is further included:

[0048] The part with a CT value less than 1000 in the tomographic image is set to zero, and the background pixels with weak response are removed from the tomographic image by adopting an OTSU threshold segmentation method.

[0049] Compared with the prior art, the present application can simultaneously obtain quality trait parameters such as fruit size, nut surface area, flesh size, fruit volume, flesh volume, fruit morphological characteristics, shell thickness, flesh proportion and mold volume through visual recognition technology, and compared with the manual trait detection method and the traditional culture method for detecting mold infection, the detection accuracy and efficiency can be greatly improved, and the visual recognition technology can realize the detection of the quality trait parameters of the nut in a nondestructive manner. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is a structural schematic diagram of a nut quality nondestructive detection device based on a full-automatic micro-CT according to an embodiment of the present application;

[0051] Figure 2 is a flowchart of a nut quality nondestructive detection method based on a full-automatic micro-CT according to an embodiment of the present application;

[0052] Figure 3 is a CUDA architecture schematic diagram of a three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm according to an embodiment of the present application;

[0053] Figure 4 is a flowchart of a tomographic image semantic segmentation method according to an embodiment of the present application;

[0054] Figure 5 is a schematic diagram of a method for extracting nut quality trait parameters according to an embodiment of the present application.

[0055] The reference numerals therein include: a flat panel detector 1, nuts 2, a micro focal spot ray source 3, a computer 4, a lifting table 5, a sample rotating table 6, a translation table 7, a ray source cooling device 8, and a PLC controller 9. DETAILED DESCRIPTION

[0056] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. In the following description, the same modules are denoted by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, detailed descriptions thereof will not be repeated.

[0057] In order to make the objectives, technical solutions, and advantages of the present application clearer, further detailed descriptions will be made to the present application in combination with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not constitute a limitation on the present application.

[0058] Figure 1 A structure of a nut quality nondestructive testing device based on a full-automatic micro-CT according to an embodiment of the present application is shown.

[0059] As shown in Figure 1 The nut quality nondestructive testing device based on the full-automatic micro-CT according to the embodiment of the present application includes the flat panel detector 1, a conveying mechanism, the micro focal spot ray source 3, the computer 4, the lifting table 5, the sample rotating table 6, the translation table 7, the ray source cooling device 8, and the PLC controller 9. The flat panel detector 1 is installed on the lifting table 5, and the height of the flat panel detector 1 is adjusted by the lifting table 5. The sample rotating table 6 and the micro focal spot ray source 3 are sequentially arranged in front of the flat panel detector 1. The nuts 2 are conveyed to the sample rotating table 6 by the conveying mechanism, and the nuts 2 are rotated by the sample rotating table 6. The sample rotating table 6 is placed on the translation table 7, and the front and back positions of the nuts 2 relative to the flat panel detector 1 are adjusted by the translation table 7. The micro focal spot ray source 3 is used for emitting X-rays, and the X-rays are received by the flat panel detector 1 after penetrating the nuts 2. The ray source cooling device 8 is installed at the bottom of the micro focal spot ray source 3, and is used for cooling the micro focal spot ray source 3 to prevent the micro focal spot ray source 3 from being damaged due to overheating. The PLC controller 9 is used for starting and stopping the conveying mechanism and the sample rotating table 6 under the control of the computer 4. The computer 4 is used for obtaining images collected by the flat panel detector 1, performing tomographic reconstruction and image processing, and obtaining quality trait parameters of the nuts.

[0060] The lifting table 5, the sample rotating table 6, and the translation table 7 are all prior art, and specific structures thereof will not be described herein.

[0061] The carrying mechanism is a mechanical hand purchased on the market. Before detection starts, the mechanical hand carries the nuts 2 from the designated position to the object carrying rotary table 6; after detection is completed, the mechanical hand carries the nuts 2 from the object carrying rotary table 6 to the designated position or other positions. Replacing manual carrying with the mechanical hand can improve efficiency, save manpower, and effectively avoid damage to the human body caused by X-rays.

[0062] The micro focal spot ray source 3 is used to provide adjustable voltage and current stable X-rays. The micro focal spot ray source 3 mainly consists of a vacuum glass tube, a cathode and an anode and the like. After the cathode filament emits electrons after heating, the electrons are bombarded at high speed to the anode target surface under the action of a high voltage electric field, and X-rays are generated.

[0063] The flat panel detector 1 is used to collect the area array image of the nuts 2. After the X-rays emitted by the micro focal spot ray source 3 penetrate the nuts 2, the X-rays reach the flat panel detector 1, the scintillation layer on the flat panel detector 1 emits electrons in a positive proportional relationship with the received X-rays, the electrons are collected by the lower layer of the silicon photodiode array, and are converted into electric charges and then into pixel values, thereby forming an area array image.

[0064] The object carrying rotary table 6 drives the nuts 2 to rotate once, and the flat panel detector 1 collects the area array image in sequence, and finally obtains the area array image of the nuts 2 at all angles, i.e. the full angle area array image.

[0065] One end of the PLC controller 9 is connected with the computer 4 to realize communication through a serial port, and the other end of the PLC controller 9 is connected with the object carrying rotary table 6 and the mechanical hand.

[0066] The computer 4 includes a device control module, an image collection module and an image processing module.

[0067] The device control module is used to send control instructions to the PLC controller and the carrying mechanism respectively, the PLC controller controls the servo motor and the driver of the object carrying rotary table based on the control instructions to realize the equal interval rotation of the nuts, and the carrying mechanism carries the nuts from the designated position to the object carrying rotary table based on the control instructions.

[0068] The image collection module is used to receive the area array image collected by the flat panel detector after the nuts rotate by one angle, and obtain the full angle area array image.

[0069] The image processing module is used to reconstruct the tomographic image based on the CUDA architecture three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm for the full angle area array image, and then perform image processing on the reconstructed tomographic image to obtain the quality trait parameters of the nuts.

[0070] The image processing module includes a tomographic image reconstruction unit, a semantic segmentation unit, a three-dimensional reconstruction unit and a quality characteristic calculation unit.

[0071] The tomographic image reconstruction unit is used for extracting the same row of the sinogram from each image of the full-angle face array image, reconstructing the tomographic image based on the three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm of the CUDA architecture, and performing noise reduction processing on the tomographic image in combination with the ray offset calibration and gain calibration.

[0072] The semantic segmentation unit is used for dividing the nut shell region, nut meat region and moldy region in the tomographic image.

[0073] The three-dimensional reconstruction unit is used for obtaining the three-dimensional digital model of the nut by sequentially stacking the tomographic image after semantic segmentation at equal steps.

[0074] The quality feature calculation unit is used for calculating the fruit size, nut surface area, meat size, fruit volume, meat volume, fruit shape feature, shell thickness, meat ratio and mold volume.

[0075] The image processing module further comprises an image preprocessing unit configured to remove the background information of the tomographic image after noise reduction.

[0076] The above describes the structure of the nut quality nondestructive detection device based on the full-automatic micro-CT provided by the embodiment of the application, and the embodiment of the application further provides a nut quality nondestructive detection method implemented by using the detection device.

[0077] Figure 2 The flow of the nut quality nondestructive detection method based on the full-automatic micro-CT provided by the embodiment of the application is shown.

[0078] As shown in Figure 2 The nut quality nondestructive detection method based on the full-automatic micro-CT provided by the embodiment of the application comprises the following steps:

[0079] S1, using a conveying mechanism to convey the nuts to the object rotating table.

[0080] S2, keeping the micro-focal spot ray source and the flat panel detector stationary, and controlling the object rotating table to rotate intermittently through the PLC controller, so as to realize the full-angle irradiation of the micro-focal spot ray source to the nuts.

[0081] S3, when the object rotating table is intermittently stopped, the flat panel detector is used to collect the face array images of the nuts at each angle, so as to obtain the full-angle face array images of the nuts and transmit them to the computer.

[0082] S4, the computer performs tomographic reconstruction and image processing based on the full-angle face array images, so as to obtain the quality trait parameters of the nuts.

[0083] In step S4, the process of performing tomographic reconstruction on the full-angle array image includes the following steps:

[0084] S41. Extract the same row from each image of the full-angle array image to form a sine graph.

[0085] S42. A high-precision decomposition CT fast reconstruction algorithm based on CUDA architecture is used to reconstruct the tomographic images, and the tomographic images are denoised by combining ray offset calibration and gain calibration.

[0086] The CUDA architecture of the high-precision decomposition CT fast reconstruction algorithm using 3D cone-beam projection is as follows: Figure 3 As shown, the formula for the high-precision decomposition CT fast reconstruction algorithm using three-dimensional cone-beam projection is:

[0087]

[0088]

[0089]

[0090] Where f(x,y,z) is the density function of the nut, β is the angle between the normal direction of the flat panel detector and the y-axis of the nut coordinate system, and R is the perpendicular distance from the origin of the nut coordinate system to the line connecting the micro-focal spot X-ray source and the projection position of the reconstruction point on the flat panel detector.

[0091] X-ray offset calibration and gain calibration are two techniques used in CT system correction to correct spatial parameters and non-uniform responses, respectively, to obtain higher quality images. The main steps of X-ray offset calibration involve quantitatively analyzing image sharpness using two-dimensional image entropy. Based on the characteristics of spatial parameters, these parameters are estimated in a specific order, and their optimal values ​​are selected based on the two-dimensional image entropy. Gain calibration primarily involves shifting the detector during imaging to stagger pixels with brightness differences and averaging the results to reduce the impact of inconsistent pixel responses.

[0092] After reconstructing the tomographic images of nuts, image processing is performed on the reconstructed tomographic images, followed by the calculation of phenotypic parameters to obtain the quality phenotypic parameters of the nuts. This invention proposes a high-precision and high-efficiency semi-automatic segmentation strategy for nut tomographic images, namely "pre-segmentation-discrimination-training-precise segmentation," to process the reconstructed tomographic images. This segmentation strategy starts from the data features of massive nut tomographic images, combining traditional digital image processing techniques with deep learning-based semantic segmentation model techniques. It avoids the problem of requiring a large amount of manual annotation for deep learning models, achieving accurate segmentation of nut tomographic images at all levels of all samples. Pre-segmentation step: Different regions of the nuts are labeled. Discrimination: High-precision tomographic images and segmentation results after pre-segmentation are manually selected as the dataset for subsequent model training. Training: A certain number of manually selected datasets are used to train the convolutional neural network model. Precise segmentation: The trained model is used to infer the segmentation of all tomographic images, and the final accurate segmentation result is obtained.

[0093] In step S4, the specific process of image processing for the tomographic image includes the following steps:

[0094] S43. Select a portion of the tomographic images as the training set, and use an image labeling method based on digital signal processing to generate different types of labels to mark the shell region, pulp region, and moldy region in the dataset.

[0095] The labeled results are used for training a deep learning semantic segmentation model based on a convolutional neural network.

[0096] Before step S43, an image preprocessing step is also included to remove background information from the noise-reduced tomographic image.

[0097] Analysis of actual imaging results revealed that when the CT value stretching window width is set to 0-5000, most air and containers, due to their extremely low radiation attenuation, will have CT values ​​below 1000. Therefore, when analyzing tomographic images, the portion with CT values ​​below 1000 is set to zero, which reduces the amount of background voxels involved in the calculation. Secondly, applying the OTSU thresholding method to the tomographic images further removes a large number of weakly responding background pixels, further suppressing background information. These two preprocessing methods can suppress the background as much as possible, reducing the amount of subsequent computation.

[0098] S44. Select labeled tomographic images with a matching degree greater than the threshold as the training set to train the deep learning semantic segmentation model based on convolutional neural networks.

[0099] S45. Perform semantic segmentation on all tomographic images using a deep learning semantic segmentation model to obtain semantic segmentation results.

[0100] After the deep learning semantic segmentation model reaches the expected accuracy, all tomographic images are input into the deep learning semantic segmentation model for semantic segmentation to obtain the final segmentation result.

[0101] Steps S43 to S45 are the flowchart of the semantic segmentation method for tomographic images, as follows: Figure 4 As shown.

[0102] Step S43 specifically includes the following steps:

[0103] S431. Analyze the different tissue characteristics of nuts from the perspectives of grayscale, gradient, and local contrast.

[0104] S432. Design feature extraction operators based on the analysis results to distinguish different tissues of nuts.

[0105] S433. Set category labels to complete the marking of the fruit shell area, fruit pulp area and moldy area.

[0106] S46. Based on the semantic segmentation results, a three-dimensional model of the nut is reconstructed to obtain a three-dimensional digital model of the nut.

[0107] Using the semantic segmentation results of tomographic images, a three-dimensional digital model of the nut is obtained by sequentially stacking images with equal step sizes.

[0108] The parsing process of the tomographic image is to store it layer by layer according to the imaging settings during scanning, in a fixed direction (vertical downward in this system) and with a fixed step size (0.1 mm in this system). Adjacent tomographic layers in the storage medium are also adjacent parts in actual space. Therefore, as long as the storage order is not changed during the image processing, the processing results are stacked sequentially along the Z-axis, and the generated three-dimensional digital model corresponds to the actual spatial information of the nut.

[0109] S47. Calculate the quality trait parameters of nuts based on a three-dimensional digital model of nuts.

[0110] The quality characteristics of nuts include nut size, nut surface area, nut volume, nut morphological characteristics, kernel size, kernel volume, shell thickness, kernel percentage, and mold volume.

[0111] The selection of quality parameters is based on the quality standards of the nut industry. Due to the variability of nut shapes and their placement during imaging, the samples need to be individualized and coded (i.e., individual samples are distinguished and numbered) before the trait calculations are performed on the samples in each region.

[0112] like Figure 5 As shown, the calculation process for the quality trait parameters of nuts is as follows:

[0113] The size of nuts is calculated as follows:

[0114] By fitting an external ellipse to the three-dimensional digital model, the major and minor axes of the minimum fitted ellipse are solved to obtain the length and width of the nut.

[0115] The calculation method for fruit pulp size is as follows:

[0116] By fitting an externally tangent ellipse to the middle section of the pulp in the three-dimensional digital model, the major and minor axes of the minimum fitted ellipse are solved to obtain the length and width of the pulp.

[0117] The method for calculating fruit pulp size only applies to the intermediate fracture layer, which can significantly improve calculation speed with almost no impact on accuracy.

[0118] The surface area of ​​nuts is calculated as follows:

[0119] The surface area of ​​the nut is calculated by analyzing the connected domains of the outermost voxels of the three-dimensional digital model.

[0120] The calculation methods for nut volume, pulp volume, and mold volume are as follows:

[0121] The volume of the fruit, pulp, and moldy parts is obtained by multiplying the set of voxels of different tissues of the nut by the actual size of each tissue voxel.

[0122] After semantic segmentation of the tomographic images, they were divided into different categories. The number of voxels in each category was directly counted to obtain the voxel set of different tissues of the nut.

[0123] The actual size of each tissue voxel is the actual spatial size of the voxel in the tomographic image, which is determined by the setting of the imaging parameters.

[0124] The calculation method for the percentage of fruit pulp is as follows:

[0125] Divide the volume of the pulp by the volume of the nut to get the proportion of pulp in the nut's volume, which is the pulp content.

[0126] The percentage of moldy plants is calculated as follows:

[0127] Divide the volume of the moldy part by the volume of the nut to get the proportion of the moldy part in the nut's volume, i.e., the mold ratio.

[0128] The thickness of the nut shell is calculated as follows:

[0129] The thickness of the nutshell is obtained by solving for the shortest distance between its inner and outer contours.

[0130] Sampling points are evenly spaced along the inner contour of the nut shell. The shortest distance from each sampling point to the outer contour is calculated, and the average value is taken as the thickness of the nut shell. The method for determining the morphological characteristics of nuts is as follows:

[0131] Nut morphological characteristics are used to determine the roundness or flatness of nuts. The closer the ratio of the length to the width of a nut is to one, the closer the nut is to being round.

[0132] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0133] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

[0134] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A non-destructive testing device for quality of nuts based on a fully automatic micro-CT, characterized by, The device comprises: a rotating platform for carrying the nuts and rotating the nuts; a conveying mechanism for conveying the nuts to the rotating platform; a translation platform for carrying the conveying mechanism and adjusting the position of the nuts; a micro-focus X-ray source for emitting X-rays to the nuts; a flat panel detector for receiving the X-rays penetrating the nuts and collecting face array images; wherein the flat panel detector collects a face array image every time the nuts rotate once, and obtains a full-angle face array image of the nuts; a lifting platform for adjusting the height of the flat panel detector; a PLC controller for starting and stopping the rotating platform and the conveying mechanism; a computer for receiving the full-angle face array image, and performing tomographic reconstruction and image processing based on the full-angle face array image to obtain quality trait parameters of the nuts; a device control module for sending control instructions to the PLC controller and the conveying mechanism, respectively; the PLC controller controls the servo motor and driver of the rotating platform based on the control instructions to realize the intermittent rotation of the nuts; and the conveying mechanism conveys the nuts from a specified position to the rotating platform based on the control instructions; an image acquisition module for receiving the face array image collected by the flat panel detector after the nuts rotate by an angle, and obtaining the full-angle face array image; an image processing module for reconstructing a tomographic image based on the full-angle face array image using a FBP algorithm, and performing image processing on the reconstructed tomographic image to obtain the quality trait parameters of the nuts; the image processing module comprises: a tomographic image reconstruction unit for extracting a same row to form a sinogram from each image of the full-angle face array image, reconstructing a tomographic image using a three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm based on a CUDA architecture, and performing noise reduction processing on the tomographic image in combination with ray offset calibration and gain calibration; wherein the formula of the three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm based on the CUDA architecture is: ； ； ； ； in, f ( x, y, z Let be the density function of the nut, and let be the angle θ between the normal direction of the flat panel detector and the y-axis of the nut's coordinate system. R The perpendicular distance from the origin of the nut coordinate system to the line connecting the projection position of the micro-focal spot ray source and the reconstruction point on the flat panel detector; a semantic segmentation unit for dividing the nut shell region, nut pulp region and moldy region in the tomographic image; wherein part of the tomographic images are selected as a training set, different types of labels are generated based on an image label generation method to mark the shell region, pulp region and moldy region in the data set, tomographic images with a matching degree greater than a threshold are selected as a training set to train a deep learning semantic segmentation model based on a convolutional neural network, and semantic segmentation is performed on all the tomographic images through the deep learning semantic segmentation model to obtain a semantic segmentation result; a three-dimensional reconstruction unit for sequentially stacking the tomographic images after semantic segmentation at equal steps to obtain a three-dimensional digital model of the nuts; a quality feature calculation unit for calculating the fruit size, nut surface area, pulp size, fruit volume, pulp volume, fruit shape feature, shell thickness, pulp ratio and moldy volume.

2. The full-automatic micro-CT based non-destructive testing device for nut quality according to claim 1, characterized in that, The image processing module further comprises an image preprocessing unit configured to remove background information of the denoised tomographic image.

3. The full-automatic micro-CT based non-destructive testing device for nut quality according to claim 1, characterized in that, The carrying mechanism is a mechanical hand.

4. A method for nondestructive testing of nut quality based on a full-automatic micro-CT, which is implemented by using the nondestructive testing device for nut quality based on a full-automatic micro-CT according to claim 1, characterized in that, The method comprises the following steps: S1, using the carrying mechanism to carry the nuts to the object rotating table; S2, keeping the micro-focus spot source and the flat panel detector static, controlling the object rotating table to rotate intermittently through the PLC controller, realizing full-angle irradiation of the micro-focus spot source to the nuts; S3, when the object rotating table is intermittently stopped, collecting the face array images of the nuts at each angle through the flat panel detector, obtaining the full-angle face array images of the nuts, and transmitting to the computer; S4, the computer performs tomographic reconstruction and image processing based on the full-angle face array images to obtain the quality trait parameters of the nuts; in step S4, the process of tomographic reconstruction of the full-angle face array images comprises the following steps: S41, extracting a sinusogram composed of the same row from each image of the full-angle face array images; S42, using a three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm based on CUDA architecture to reconstruct the tomographic images, and combining with ray offset calibration and gain calibration to perform denoising processing on the tomographic images; wherein, The formula of the three-dimensional cone beam projection high-precision decomposition CT fast reconstruction algorithm based on CUDA architecture is: ； ； ； ； wherein, f (0.5 < a < 1) is a constant, x, y, z is a density function of the nut, an angle β between a normal direction of the flat panel detector and a y-axis of a nut coordinate system, R is a vertical distance from a coordinate origin of the nut coordinate system to a line connecting the micro focal spot ray source and a projection position of the reconstruction point on the flat panel detector; In step S4, the process of image processing of the tomographic images comprises the following steps: S43, selecting part of the tomographic images as a training set, and using an image label generation method to generate different types of labels to mark the shell region, pulp region and moldy region in the data set; S44, selecting the tomographic images with labels having a matching degree greater than a threshold value as a training set to train a deep learning semantic segmentation model based on convolutional neural network; S45, performing semantic segmentation on all the tomographic images through the deep learning semantic segmentation model to obtain a semantic segmentation result; S46, based on the semantic segmentation result, reconstructing a three-dimensional model of the nuts to obtain a three-dimensional digital model of the nuts; S47, based on the three-dimensional digital model of the nuts, calculating the quality trait parameters of the nuts.

5. The non-destructive quality detection method of nuts based on fully automatic micro-CT according to claim 4, characterized in that, Step S43 specifically comprises the following steps: S431, analyzing different tissue features of the nuts from the angles of gray scale, gradient and local contrast; S432, designing a feature extraction operator according to the analysis result to distinguish different tissues of the nuts; S433, setting a class label to complete the marking of the shell region, pulp region and moldy region.

6. The non-destructive quality detection method of nuts based on a fully automatic micro-CT according to claim 4 or 5, characterized in that, In step S46, the three-dimensional digital model of the nuts is obtained through equal-step sequential stacking using the semantic segmentation result of the tomographic images.

7. The full-automatic micro-CT based non-destructive quality detection method of nuts according to claim 4, characterized in that, Before step S43, an image preprocessing step is further included: Setting the part with CT value less than 1000 in the tomographic image to zero, and cutting off the background pixels with weak response from the tomographic image by using an OTSU threshold segmentation method.

Images