Egg meat spot nondestructive testing method based on nuclear magnetic resonance imaging technology

Abstract

The invention relates to an egg blood and meat spot nondestructive testing method based on a nuclear magnetic resonance imaging technology. The method comprises the following steps: (1) establishing an egg sample set; (2) collecting; egg image sample data; (3) performing batch processing on a DICOM file of the acquired egg image sample data; (4) feature extraction of the egg meat spot image sample; (5) identifying and modeling meat spot and frenum areas; and (6) meat spot area identification and identification. According to the method, the MRI technology is used for non-destructive testing of the meat spots in the eggs, different relaxation signal intensities are displayed through different components in the eggs in MRI scanning, substances such as egg yolk, egg white and the meat spots are clearly recognized, non-destructive testing of the meat spots with the diameter being 2 mm or above in the eggs and positioning of the occurrence positions of the meat spots are achieved, the detection accuracy is improved, and the detection cost is reduced. Internal information can be obtained without breaking the eggs, incubation can be continued, the detection cost is reduced, and the method has remarkable practical guidance value for egg quality improvement and laying hen breeding.

Description

Technical Field

[0001] The invention relates to the field of poultry egg detection, and in particular to a non-destructive detection method for egg meat spots based on nuclear magnetic resonance imaging technology. Background Art

[0002] Blood spots on eggs are blood spots or blood clots in the yolk and egg white, while meat spots refer to meat-like tissues with different shades of pink to brown in the yolk and egg white. The presence of blood spots and meat spots affects the quality of eggs, and thus affects consumers' purchasing desire and the sales price of eggs, causing potential economic losses to poultry and egg companies.

[0003] As far as the detection of egg meat spots is concerned, poultry egg companies currently rely mainly on manual sorting, which is labor-intensive, low-efficiency, and the detection accuracy cannot be guaranteed; and the eggs need to be broken for testing, and the broken eggs cannot be sold or hatched again. Therefore, this method is not suitable for large-scale batch application.

[0004] Non-destructive testing is the development direction of poultry and egg testing. Combined with advanced image processing technology and automated testing equipment, it can significantly improve the efficiency and accuracy of poultry and egg testing.

[0005] In recent years, detection methods based on image feature acquisition and machine learning have been preliminarily studied. The existing technology has developed a multi-view stereoscopic vision (MVS) system that uses strong light to transmit eggs to obtain egg images, and has developed an algorithm to detect egg blood spots. However, this method is easily affected by external factors. For example, factors such as dirt on the surface of the egg will interfere with the egg image and cause misjudgment. Furthermore, changes in factors such as temperature and humidity will also affect the performance of the image sensor and cause misjudgment.

[0006] Spectroscopic technology and hyperspectral imaging (HIS) have developed rapidly in the field of nondestructive testing in recent years. Existing technologies have developed a visible light-near infrared (Vis-NIR) system consisting of a handheld spectrometer, optical fiber and halogen lamp to detect blood spots in brown eggs. The two methods are limited by laboratory scale and cannot achieve the processing speed required for commercial applications. In addition, the detection instruments are expensive and have high maintenance costs, making it difficult to popularize them to small and medium-sized enterprises or ordinary farms. In addition, current research is limited to the detection of blood spot eggs, while ignoring the detection of meat spot eggs.

[0007] Magnetic Resonance Imaging (MRI) is a non-destructive testing technology that uses static magnetic fields and radio frequency magnetic fields to image tissues. It belongs to water proton imaging, which uses an external magnetic field to change the spin direction of electrons around water protons, and uses this principle to produce proton imaging. During the imaging process, neither electron ionizing radiation nor contrast agents are used to obtain high-contrast clear images. MRI technology can directly display the coronal, sagittal, transverse and any other directional images of the scanned object, with high resolution, clear images and details. Only the effects of static magnetic fields and radio frequency magnetic fields on water protons affect the image, and it is not easily affected by external factors. There have been no reports on the application of magnetic resonance technology to the detection of blood spots or meat spots on eggs. Summary of the invention

[0008] The present invention aims to solve the technical problems existing in the prior art and provide a nondestructive detection method for egg meat spots based on nuclear magnetic resonance imaging technology.

[0009] The method uses MRI technology to non-destructively detect meat spots in eggs. Different components inside the egg show different relaxation signal intensities in MRI scans, so that substances such as yolk, egg white, and meat spots can be clearly identified, and non-destructive detection of meat spots with a diameter of more than 2 mm in eggs and the location of the meat spots are achieved. The accuracy of the detection is improved, and the internal information of the egg can be obtained without breaking the egg. The egg can continue to be incubated, which reduces the detection cost and has significant practical guidance value for improving egg quality and laying hen breeding. In order to achieve the purpose of the invention, the technical solution provided by the present invention is as follows:

[0010] A non-destructive detection method for blood and flesh spots of eggs based on nuclear magnetic resonance imaging technology comprises the following steps:

[0011] (1) Creating an egg sample set: This includes dividing the egg samples into a training set and a test set. The training set is used for model training, and the test set is used to evaluate the prediction accuracy of the model.

[0012] 60 unfertilized eggs of any breed were selected as training set samples; 51 unfertilized eggs were selected as test samples; the eggshell colors included white, pink, and brown;

[0013] Make the egg yolk meat spot model: Use a cutter to cut open the blunt end of the eggshell. If there is a natural meat spot in the egg, use a ruler to measure the size of the meat spot. If there is no natural meat spot in the egg, take 3-4 egg yolks, take the egg yolk liquid and merge it into an eggshell to simulate a full egg yolk environment. Put in small round breast meat pieces with diameters of 4mm, 3mm, 2mm, and 1mm, and seal with a sealing film.

[0014] To make an egg white meat spot model: make a pinhole-sized hole at the blunt end of the eggshell, insert a syringe obliquely into the egg white, and slowly poke the meat sample with the needle without sealing it.

[0015] (2) Collect egg image sample data:

[0016] Collecting egg image sample data includes: using high magnetic field nuclear magnetic field scanning and nuclear magnetic resonance imaging, using head and neck combined coil for nuclear magnetic field scanning, using multivariate parameter acquisition method to collect egg internal image sample data, and outputting the collected image sample data in the form of DICOM file.

[0017] The field strength of the high magnetic field nuclear magnetic resonance scanning is 3.0T.

[0018] The main parameters involved include: window width / window level (WW / WL), repetition time (TR), echo time (TE), slice thickness / slice spacing (Thk / p), field of view (FOV), number of slices (Slices), voxel size, etc.

[0019] Preferably, the window width / window position (WW / WL) is: 337 / 152; the repetition time (TR) is 1300 ms, the echo time (TE) is 246.96 ms; the layer thickness / layer spacing (Thk / p) is 1 mm / 0.6 mm; the field of view (FOV) is 300.0 mm*212.0 mm; the number of slices is 166, and the voxel size is 0.59*0.59*1 mm.

[0020] (3) Batch processing of the DICOM files of the acquired egg image sample data

[0021] Batch processing of the DICOM files of the acquired egg image sample data includes:

[0022] The 3D slicer tool was used to convert the DICOM file of the egg image layer by layer to obtain the cross-sectional view of the egg image sample; the 3D Slider tool was used to obtain the summary output folder containing the DICOM file converted into images frame by frame; in the collected egg image samples, the egg white was a white image, the foreign matter in the egg white was black, the yolk was a black image, and the foreign matter in the yolk was grayish white; in the egg white, at both ends of the yolk, a band or mass tissue (i.e., frenulum) could be seen, the frenulum itself was white, and appeared grayish black in the image.

[0023] Then, the ImageJ tool was used to cut the egg sample images into monomer images to obtain cross-sectional images for subsequent image analysis.

[0024] (4) Feature extraction of egg spot image samples

[0025] Including: using the ImageJ tool to open the image to be measured, and selecting Image-Type-8bit in the toolbar, and using a linear scaling algorithm to convert the egg image sample into an 8-bit image;

[0026] Select Image-Duplicate in the toolbar, copy the original image 3 times, and number them as samples a, b, and c. Select Image-Type-8bit in the toolbar to convert the RGB color space of the 3 image samples to 8bit in preparation for subsequent processing;

[0027] Select Process-Filiters-Gaussian Blur in the toolbar, perform Gaussian blur algorithm processing on image sample b, and set the σ value to 4;

[0028] Subtract the Gaussian blurred sample b from the sample a using a subtraction operation, perform a multiplication operation on the subtraction result, fix the enhancement value to 20, and obtain the internal contour feature image of the egg internal image sample;

[0029] The obtained egg internal contour feature image is added to the graphic sample c to obtain an egg image sample with the internal contour enhanced. The sample is subjected to 8-bit operation for automatic identification of the internal area of ​​the egg.

[0030] (5) Identification and modeling of flesh spots and frenulum areas

[0031] Including: importing the obtained 8-bit image samples into the Trainable Weka Segmentation platform, inputting them into the Weka algorithm, and using Gaussian blur, Hessian matrix, sobel filter, Gaussian difference and membrane projection to collect feature data;

[0032] Use Create new class to create a new class. The default number of classes in the plug-in is two. This operation will increase the number of classes to n, where n depends on the number of categories to be distinguished, and change the name of the new class in the "Settings" dialog box; Create Model: select some feature images of egg white spots and egg yolk spots in turn and classify them into class 2 and class 3 respectively, select feature images of normal background areas such as eggshell, egg white, egg white, and lace and classify them into class 1.

[0033] In order to improve the accuracy of meat spot recognition under MRI scanning, two models can be created. Model 1: select part of the lace feature image and classify it into class 2, select normal eggshell, egg white, egg white, meat spots, etc. as background area feature images and classify them into class 1; Model 2: select part of the egg white meat spots, egg yolk meat spots, and lace feature images in turn and classify them into class 2, class 3, and class 4 respectively, select normal eggshell, egg white, egg white and other background area feature images and classify them into class 1.

[0034] The training process is activated using Train classifier, which guarantees at least two classes. The features of the input image are extracted and converted into a set of floating point value vectors according to the format expected by the Weka classifier. The running time depends on the size of the image, the number of features, and the number of cores of the computer running Fiji. The feature calculation is done in a fully multithreaded manner.

[0035] Use Create result to create and display the resulting image. This image is equivalent to the current overlay (8-bit color with the same class color), with each pixel set to the index value of the most likely class (0, 1, 2, ...).

[0036] Use Save classifier to store the current classifier for later use. Use Save data to save the current trace information to a data file, which can be processed later using the plugin or Weka Explorer itself. The plugin will save the feature vectors derived from the pixels belonging to each trace to an ARFF file at a location selected by the user.

[0037] To ensure the accuracy of the classifier results, you can use Load data to continuously expand the training set and repeat the training. The more detailed and richer the training set, the more accurate the trained classifier results.

[0038] Use Load classifier to load the classifier saved in the previous step. The plugin will use the properties of this new classifier to check and adjust the selected features. The classifier file is in the standard Weka format (.model). Use Load data to load data (Weka format) from the same or other images or previous traces on the stack. Similarly, the plugin will check and enforce consistency between the loaded data and the current image, features, and classes. The input file format is the standard Weka format: ARFF.

[0039] Click Create result to generate the result graph and save it in a folder.

[0040] (6) Identification of meat spot areas

[0041] Convert the result obtained in step (5) to an 8-bit image. Click Process-Binary-Convert to Mask in the toolbar to generate a binary image for analysis.

[0042] Click Process-Binary-Fill Holes in the toolbar to fill the gaps in the image for easier calculation;

[0043] Click Image-Adjust-Threshold in the toolbar, adjust the threshold range using the slider, and select the meat spot area;

[0044] Click Analyze-Analyze Particles in the toolbar to automatically analyze, count foreign particles, and perform individual area statistics, and export the results;

[0045] In the statistical results, eggs with a circled area of ​​22-37 pixels are judged as meat spot eggs based on the comparison between the actual size of the meat spot and the image, and the rest are judged as normal eggs. Optionally, the results of Model1 and Model2 are subtracted, and the Model1 results with a value less than 50 pixels are eliminated, and then all negative differences are eliminated, and the differences between 0-200 pixels are screened out to determine that meat spots exist. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 : Comparison chart of scanning results of MRI technology and CT technology of the present invention. DETAILED DESCRIPTION

[0047] The present invention has been introduced in detail above. Specific examples have been used in the specification to illustrate the principles and implementation methods of the present invention. However, the examples are only used to help understand the present invention and should not be construed as limiting the present invention.

[0048] Example 1: Detection of meat spots on eggs using 3.0T MRI

[0049] 60 eggs were used as training set samples, of which 6 were meat-spotted eggs. After 3.0TMRI scanning, 11 eggs were identified as meat-spotted eggs.

[0050] The eggs were broken for verification, and all 6 real meat spot eggs were tested, and 5 normal eggs were judged to be meat spot eggs.

[0051] The total recognition rate of the training set samples (number of normal eggs recognized / total number of eggs) is 91.6%.

[0052] A total of 51 eggs were used as test set samples. After 3.0TMRI scanning, 9 eggs were identified as meat spot eggs. After breaking the eggs for verification, all 4 real meat spot eggs were detected, and 5 normal eggs were identified as meat spot eggs.

[0053] The total recognition rate of the test samples (number of normal eggs identified / total number of eggs) was 90.1%.

[0054] According to the experimental results, the present invention can identify meat spot eggs in batches, and the minimum limit for identifying the size of the meat spots is 2 mm, which meets the maximum limit required by GB / T 39438-2020 "Packaged Eggs".

[0055] Comparative Example 1: Traditional Manual Detection Method

[0056] The traditional manual detection method for detecting meat spots in eggs mainly relies on human visual observation and experience judgment, which is easily affected by human factors. At the same time, it lacks data recording and tracking, and cannot carry out comprehensive quality control.

[0057] In addition, the manual detection method is cumbersome and requires breaking the eggs for detection. Broken eggs cannot be sold as whole eggs, especially for breeding eggs, as broken eggs lose the conditions for hatching.

[0058] Manual inspection also faces challenges such as high labor costs, long training cycles and safety risks, which limit the large-scale application of manual inspection methods.

[0059] Comparative Example 2: Detection of egg spots using CT scanning

[0060] CT imaging uses X-ray beams to scan the object to be observed. In medicine, X-rays are used to perform tomographic scanning on a certain part of the human body to obtain a cross-sectional or stereoscopic image of the examined part of the human body. It can provide complete three-dimensional information of the examined part of the human body and can clearly show the organs and structures.

[0061] CT belongs to density imaging, which uses the density difference of the internal structure of the target to be measured. The density difference leads to different absorption rates of radiation, and the density difference image is generated through post-processing. Therefore, CT can only distinguish tissues with density differences, and the resolution of soft tissues is not high.

[0062] Since the density of the material inside the egg is very similar, the CT scanning effect is very unsatisfactory. In addition, since CT uses ionizing radiation for imaging, the radiation has a certain risk of teratogenicity, which may have an adverse effect on the hatching of the eggs.

[0063] This experiment used CT computed tomography, 1.5T low-field magnetic resonance, and 3.0T high-field magnetic resonance to scan the internal structure of eggs. The scan results are shown in Figure 1 . Figure 1In the figure, a represents the image obtained by CT scanning; b represents the image obtained by 1.5T low-field strength magnetic resonance technology; c represents the image obtained by 3.0T high-field strength magnetic resonance technology.

[0064] Depend on Figure 1 It can be seen that due to the small density difference of the material inside the egg, CT cannot be used to detect foreign matter in the egg. Magnetic resonance imaging is suitable for detecting soft tissues, with high imaging contrast. When the field strength reaches 3.0T, the meat spots inside the egg can be clearly identified.

[0065] Technical effects of the present invention:

[0066] (1) The present invention uses MRI technology for the first time to achieve non-destructive detection of meat spots in eggs. Different components inside the egg show different relaxation signal intensities in MRI scans, so that the yolk, egg white, meat spots and other substances in the egg can be clearly identified. The egg can be obtained without being affected by the shell color, and the egg can continue to be incubated without being broken, which reduces the detection cost and provides a new method and idea for egg quality detection.

[0067] (2) The present invention uses 3T MRI combined with a machine learning random forest algorithm to achieve non-destructive detection of meat spots with a diameter of more than 2 mm inside eggs. The location of the meat spots can be accurately located based on the relative position of the meat spots, yolk, and eggshell in the image. The recognition success rate of the present invention is more than 90%, which is significantly higher than that of traditional detection methods.

[0068] (3) The present invention has the characteristics of high precision and non-destructive detection. The use of this high-precision non-destructive blood and flesh spot detection technology can greatly improve the quality of eggs and has significant practical guidance value for laying hen breeding.

[0069] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A non-destructive detection method for blood and flesh spots of eggs based on nuclear magnetic resonance imaging technology, characterized in that: The detection method comprises the following steps: (1) Create an egg sample set: Divide the egg samples into a training set and a test set. The training set is used for model training, and the test set is used to evaluate the prediction accuracy of the model. (2) Collecting egg image sample data: using high magnetic field nuclear magnetic resonance scanning and nuclear magnetic resonance imaging, using a multivariate parameter acquisition method to collect egg internal image sample data, and the collected image sample data is output as a DICOM file; (3) Batch processing the DICOM files of the acquired egg image sample data: using the 3Dslicer tool to convert the DICOM files of the egg images layer by layer to obtain the cross-sectional images of the egg image samples; using the 3D Slider tool to obtain a summary output folder containing the DICOM files converted into images frame by frame; in the collected egg image samples, the egg white is a white image, the foreign matter in the egg white is black, the egg yolk is a black image, and the foreign matter in the egg yolk is grayish white; in the egg white, there is a band or mass tissue at each end of the yolk, namely the frenulum, and the frenulum is white; Based on the ImageJ tool, the collective image of the egg was cut into individual images to obtain the cross-sectional image, and the egg sample image was cut into individual images using the ImageJ tool; (4) Feature extraction of egg meat spot image samples: Use ImageJ tool to open the image to be tested, select Image-Type-8bit in the toolbar, and use linear scaling algorithm to convert the egg image sample into an 8-bit image; (5) Identification and modeling of flesh spots and frenulum areas: The obtained 8-bit image samples were imported into the Trainable WekaSegmentation platform and input into the Weka algorithm, and feature data were collected using Gaussian blur, Hessian matrix, sobel filter, Gaussian difference, and membrane projection; (6) Identification of meat spots: The result obtained in step (5) is converted into an 8-bit image, and a binary image is generated for analysis to determine the meat spots in the eggs.

2. The nondestructive testing method according to claim 1, characterized in that: Step (1) creating an egg sample set, including: 60 unfertilized eggs of any breed were selected as training set samples; 51 unfertilized eggs were selected as test samples; the eggshell colors included white, pink, and brown; Make the egg yolk meat spot model: Use a cutter to cut open the blunt end of the eggshell. If there is a natural meat spot in the egg, use a ruler to measure the size of the meat spot. If there is no natural meat spot in the egg, take 3-4 egg yolks, take the egg yolk liquid and merge it into an eggshell to simulate a full egg yolk environment. Put in small round breast meat pieces with diameters of 4mm, 3mm, 2mm, and 1mm, and seal with a sealing film. To make an egg white meat spot model: make a pinhole-sized hole at the blunt end of the eggshell, insert a syringe obliquely into the egg white, and slowly poke the meat sample with the needle without sealing it.

3. The nondestructive testing method according to claim 1, characterized in that: Step (2) collecting egg image sample data includes: using 3.0T high magnetic field nuclear magnetic resonance scanning and performing nuclear magnetic resonance imaging, using a head and neck combined coil for nuclear magnetic resonance scanning, and using a multivariate parameter acquisition method to collect egg internal image sample data, and the collected image sample data is output as a DICOM file; the parameters involved include: window width / window level (WW / WL), repetition time (TR), echo time (TE), layer thickness / layer spacing (Thk / p), field of view (FOV), number of slices (Slices) and voxel size.

4. The nondestructive testing method according to claim 1, characterized in that: Step (4) extracting features of the egg meat spot image sample comprises: using the ImageJ tool to open the image to be tested, selecting Image-Type-8bit in the toolbar, and converting the egg image sample into an 8-bit image by using a linear scaling algorithm; selecting Image-Duplicate in the toolbar to copy 3 copies of the original image and number them as samples a, b, and c in sequence; and selecting Image-Type-8bit in the toolbar to convert the 3 image samples from RGB color space to 8-bit; selecting Process-Filiters-Gaussian Blur in the toolbar to perform Gaussian blur algorithm processing on image sample b, and setting the σ value to be fixed at 4; using a subtraction operation to subtract the Gaussian blurred image of graphic sample a from the image of graphic sample b, performing a multiplication operation on the subtraction operation result, fixing the enhancement value to 20, and obtaining an internal contour feature image of the egg internal image sample; performing an addition operation on the obtained egg internal contour feature image and graphic sample c to obtain an egg image sample with an enhanced internal contour, and performing an 8-bit operation on the sample so as to automatically identify the internal area of ​​the egg.

5. The nondestructive testing method according to claim 1, characterized in that: Step (5) identification and modeling of flesh spots and frenulum areas also includes the following operations: Use Create new class to create a new class. The default number of classes in the plugin is two. This operation increases the number of classes to n, where n depends on the number of categories to be distinguished. Change the name of the new class in the "Settings" dialog box. Create Model: Select some of the egg white spots and egg yolk spots feature images and classify them into class 2 and class 3 respectively, select the normal eggshell, egg white, egg white, lace and other background area feature images and classify them into class 1; Or create two models, where Model 1: select part of the lace feature image and classify it into class 2, select normal eggshell, egg white, egg white, meat spots, etc. as background area feature images and classify them into class 1; Model 2: select part of the egg white meat spots, egg yolk meat spots, and lace feature images in turn and classify them into class 2, class 3, and class 4 respectively, select normal eggshell, egg white, egg white, etc. as background area feature images and classify them into class 1; Use Train classifier to activate the training process; use Create result to create and display the generated image; Use Save classifier to store the current classifier; Use Load classifier to load the classifier saved in the previous step; Generate result graphs and save them in a folder.

6. The nondestructive testing method according to claim 5, characterized in that: Step (6) identifying the meat spot area includes: Convert the result obtained in step (5) to an 8-bit image. Click Process-Binary-Convert to Mask in the toolbar to generate a binary image for analysis. Click Process-Binary-Fill Holes in the toolbar to fill the gaps in the image for easier calculation; Click Image-Adjust-Threshold in the toolbar, adjust the threshold range using the slider, and select the meat spot area; Click Analyze Particles in the toolbar to automatically analyze, count foreign particles, and perform individual area statistics, and then export the results; When a model is created, in the statistical results, eggs with an area of ​​22-37 pixels are judged as meat spot eggs based on the actual size of the meat spot and the image, and the rest are judged as normal eggs; When two models are created, the difference between the results of Model 1 and Model 2 is calculated, and the Model 1 results with a value less than 50 pixels are eliminated. Then all negative differences are eliminated, and the results with a difference between 0-200 pixels are screened out to determine that meat spots exist.

7. The nondestructive testing method according to claim 3, characterized in that: Window width / window level (WW / WL): 337 / 152; repetition time (TR): 1300ms, echo time (TE): 246.96ms; slice thickness / slice spacing (Thk / p): 1mm / 0.6mm; field of view (FOV): 300.0mm*212.0mm; number of slices: 166 slices; voxel size: 0.59*0.59*1mm.

Images