Method for positioning and detecting key parts of hatching eggs

Abstract

The invention discloses a method for positioning and detecting key parts of hatching eggs. The method comprises the following steps: acquiring transmission images in hatching eggs of different days through a hatching egg image acquisition device, and training and establishing a hatching egg physiological state judgment model and a hatching egg key part segmentation model; the discrimination model outputs a transmission image of the fertilized egg to be positioned and detected, the key part is identified after the transmission image is processed by the segmentation model, then the optimal coordinate point extracted by the allantoic fluid is positioned by adopting an image processing method, and the allantoic fluid is extracted to realize the detection of the hatching egg. According to the method, a dual model of hatching egg development evaluation and key part identification is established, interference of human subjective factors is weakened, and detection and classification errors are reduced; the allantoic fluid coordinate positioning method reduces the risk of puncturing and damaging the embryo and improves the allantoic fluid extraction success rate; according to the invention, rapid and accurate hatching egg information evaluation and male and female chick embryo difference characteristic detection can be provided according to hatching application requirements.

Description

Technical Field

[0001] The invention relates to a method for detecting a breeding egg, relates to detection technology in the poultry egg industry and the field of computer vision, and in particular to a method for positioning and detecting key parts of a breeding egg. Background Art

[0002] The detection and evaluation of breeding eggs is the core link of the poultry breeding and hatching industry, which is directly related to the utilization rate of breeding eggs, hatching efficiency and subsequent production benefits. Traditional breeding egg detection mainly relies on manual egg illumination and sampling destructive detection, which has problems such as low efficiency, high misjudgment rate and large breeding egg loss. For example, manual inspection is not only labor-intensive, but also subjective in judging the embryonic development status (such as infertile eggs, dead sperm eggs, fertilized eggs), which is difficult to meet the needs of large-scale breeding. In the prior art, although there is a breeding egg inspection device based on optical transmission (such as obtaining a transmission image through a light source and a light receiving unit), it relies on a fixed light source layout, and the positioning accuracy of key parts of the breeding eggs (such as embryos, air chambers, and blood vessels) is insufficient, and it lacks the ability to automatically classify multiple categories of physiological states, resulting in limited reliability of the test results. In addition, the sex identification, developmental evaluation and quality detection of breeding eggs mostly rely on molecular biology techniques or destructive sampling after hatching, which not only lags behind the production cycle, but also causes a waste of resources. In recent years, non-destructive testing technologies (such as visible near-infrared spectroscopy and Raman spectroscopy) have been tried for egg quality analysis, but they are limited by inaccurate positioning of key parts and high complexity of data modeling, making it difficult to achieve efficient and accurate online testing. With the development of deep learning and semantic segmentation technology, how to build a dedicated model that adapts to the multi-scale characteristics of eggs and achieve accurate positioning and non-destructive testing of key parts has become a technical bottleneck that the industry urgently needs to break through. Summary of the invention

[0003] In order to solve the problems existing in the background technology, the present invention provides a method for locating and detecting key parts of breeding eggs.

[0004] The technical solution adopted by the present invention is: The method for positioning and detecting key parts of a breeding egg of the present invention comprises: Step 1: Build a breeding egg image acquisition device to collect transmission images of the inside of several breeding eggs of different days, and obtain a breeding egg physiological state discrimination model and a breeding egg key part segmentation model after training through a deep learning classification model and a semantic segmentation model.

[0005] Step 2, using a breeding egg image acquisition device to acquire a transmission image of the inside of the breeding egg to be located and detected, inputting the transmission image of the breeding egg to be located and detected into a breeding egg physiological state discrimination model, and obtaining a transmission image of the identified fertilized egg after processing.

[0006] Step 3, inputting the identified transmission image of the fertilized egg into the key part segmentation model of the breeding egg, and identifying the key parts of the breeding egg to be located and detected after processing, thereby obtaining the transmission image of the segmented fertilized egg.

[0007] Step 4: Use image processing methods to locate the optimal coordinate points for allantoic fluid extraction from the transmission image of the segmented fertilized egg, extract the allantoic fluid at the optimal coordinate points, and detect the breeding eggs through the extracted allantoic fluid. Use spectral non-destructive testing technology or molecular biology minimally invasive testing technology to detect the information of the breeding eggs for subsequent sex identification, development evaluation and quality testing of the breeding eggs.

[0008] In the step 1, the breeding egg image acquisition device includes an image collector, a light source, an egg tray, an egg candling device, a rotating bracket, a shielding ring and a dark box, the image collector, the egg tray, the egg candling device, the rotating bracket and the shielding ring are all located in the dark box, the egg tray is horizontally installed at the center bottom of the rotating bracket, the breeding eggs are vertically placed on the egg tray with the blunt end facing upward, the egg candling device is installed at the center top of the rotating bracket and the bottom end is located directly above the breeding eggs, the top of the breeding eggs and the bottom end of the egg candling device are sealed by a shielding ring that isolates light, the light source is installed inside the egg candling device and the light source is vertically downward toward the breeding eggs directly below, the image collector is installed on the side of the rotating bracket and the lens is horizontally toward the breeding eggs; when the breeding egg image acquisition device acquires images, the breeding eggs are irradiated by the light source, the bracket is rotated, and the rotating bracket is controlled to drive the image collector to rotate 360° around its own vertical center axis, and after setting the image acquisition parameters, the internal transmission images of the breeding eggs in all directions are acquired by the image collector, and then transmitted to the computer for storage.

[0009] In step 1, the breeding eggs are all eggs that have been incubated for 3-15 days.

[0010] In the step 1, different categories of breeding eggs include unfertilized eggs, dead sperm eggs and fertilized eggs; after adding category labels to the transmission images inside various breeding eggs, the deep learning model is trained to obtain a breeding egg physiological state discrimination model, so as to discriminate the category of breeding eggs according to the breeding egg physiological state discrimination model, obtain the transmission image of the fertilized egg, and then carry out the next step of positioning and detection.

[0011] In the step 3, the transmission images of the identified fertilized eggs are screened to select the transmission images of the fertilized eggs with embryos, blood vessels and air chambers, and the areas of the embryos, blood vessels and air chambers in each screened image are annotated, and then input into the semantic segmentation model for training to obtain a segmentation model of key parts of the breeding eggs.

[0012] In the step 4, the transmission image of the segmented fertilized egg is used to obtain the optimal coordinate points for allantoic fluid extraction using an image processing method. First, an image coordinate system is established on the transmission image of the segmented fertilized egg in the horizontal and vertical directions to obtain the contours and position coordinates of the embryo, blood vessels and air chambers. The contour of the embryo is expanded equidistantly to obtain the embryo expansion area. The embryo expansion area and the regional contour of the blood vessels are compared. If there is a repeated area in the contour, all coordinate points in the repeated area of ​​the contour are set as repeated coordinate points, and the repeated coordinate points are removed. The remaining coordinate points in the contour of the embryo expansion area and the coordinates in the contour of the air chamber area are traversed, and the Euclidean distance is calculated. The tree-shaped KDTree (k-dimensional) data structure is used to accelerate the nearest neighbor search to obtain the shortest Euclidean distance, so that the coordinate points in the contour of the embryo expansion area are located according to the shortest Euclidean distance as the optimal coordinate points for allantoic fluid extraction and are marked.

[0013] In step 4, the extracted allantoic fluid of the breeding eggs is spectrally scanned to obtain spectral data for testing the breeding eggs.

[0014] Based on spectral technology, non-destructive testing can be performed on key parts of the eggs, such as the allantois, embryos, and blood vessels. Combined with machine learning, a non-destructive testing model can be constructed for sex identification, egg development assessment, egg quality testing, etc. Spectral technology includes visible near-infrared spectroscopy, Fourier transform near-infrared spectroscopy, Raman spectroscopy, infrared spectroscopy, and terahertz spectroscopy.

[0015] The key part positioning and detection system of the breeding egg of the present invention comprises: The data acquisition processing unit is used to acquire the transmission image inside the breeding egg, and obtain the transmission image of the fertilized egg after identification after processing by the breeding egg physiological state identification model and the breeding egg key part segmentation model.

[0016] The key part positioning and detection unit is used to obtain the segmented transmission image of the fertilized egg after processing the transmission image of the fertilized egg through the key part segmentation model of the breeding egg, and then use the image processing method to locate the optimal coordinate point for allantoic fluid extraction, and extract the allantoic fluid according to the optimal coordinate point for subsequent detection.

[0017] The electronic device of the present invention comprises: a memory and a processor coupled to each other, wherein the memory stores program data, and the processor calls the program data to execute the method as described above.

[0018] The computer-readable storage medium of the present invention stores program data thereon, and when the program data is executed by a processor, the method described above is implemented.

[0019] The beneficial effects of the present invention are: The present invention combines the deep learning classification model with the semantic segmentation model to realize the automatic identification of the physiological state of the eggs (infertile eggs / dead sperm eggs / fertilized eggs) and the accurate segmentation of key parts (embryos, air chambers and blood vessels, etc.). Compared with the traditional manual egg candling technology, it significantly improves the identification accuracy (especially for the instability of embryonic development) and reduces the subjective error of human beings.

[0020] The present invention integrates the developmental characteristics of breeding eggs with image processing algorithms to design an optimal coordinate positioning method for allantoic fluid extraction. The method reduces the risk of puncture damage to the embryo and improves the success rate of allantoic fluid extraction and the hatching rate by excluding overlapping blood vessels and optimizing spatial distance calculation.

[0021] The information on hatching eggs obtained by the present invention can be applied to multimodal hatching application requirements such as hatching egg sex identification, development assessment and quality inspection. Compared with a single detection method, the system combines the physical characteristics of image positioning with the biochemical characteristics of allantoic fluid, providing double verification for early hatching evaluation of hatching eggs. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flowchart of the method of the present invention; Figure 2 It is a schematic diagram of the image acquisition device of the present invention; Figure 3 This is a diagram of the physiological characteristics of the breeding eggs of the present invention; Figure 4 It is a schematic diagram of the allantois positioning process of the present invention; Figure 5 The present invention is part of the experimental results of the allantoic fluid extraction rate and the chicken embryo hatching rate; In the figure: 1. breeding eggs, 101. unfertilized eggs, 102. dead sperm eggs, 103. fertilized eggs, 104. first development breeding eggs, 105. second development breeding eggs, 106. third development breeding eggs, 107. fourth development breeding eggs, 108. allantois, 109. embryo, 110. blood vessels, 111. air chamber, 112. embryo expansion area, 115. embryo expansion area coordinate points, 2. image collector, 3. light source, 4. egg tray, 5. egg illuminator, 6. rotating bracket, 7. shielding ring. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Egg candling is also called egg testing, which is to use the light of the egg candling device 5 to see the development of the embryo 109 and the internal quality of the egg 1. Because there are many pores on the surface of the egg shell, the distribution is uneven (300-370 pores / cm at the blunt end and 300-370 pores / cm at the blunt end). 2 , tip 150-180 pieces / cm 2), of different sizes (9×10-22×29 μm), making the egg transparent. The pores can communicate the inside and outside of the egg 1 to ensure the normal development of the embryo 109. After a certain period of incubation, the contents of the egg 1 change to form different quality conditions. Under dark conditions, the egg light 5 is used to see through the egg 1, and the eggshell, the height of the air chamber 111, the egg white, the yolk, the ligament and the embryo 109 can be observed to identify the quality of the egg 1 and make a comprehensive assessment. The principle of the light transmission method is: according to the pores on the eggshell of poultry eggs that can transmit light, the contents of the egg 1 change to form different quality conditions. Under the light perspective, the eggshell, the height of the air chamber 111, the egg white, the yolk, the ligament and the embryo 109 can be observed to identify the quality of the egg 1 and make a comprehensive assessment. Using the light transmission method combined with computer vision to monitor and classify the physiological development characteristics of the egg 1 can assist in locating the key parts of the egg 1. Based on the above principle, the present invention proposes a method for positioning and detecting key parts of a breeding egg, so as to realize the classification of the physiological state of the breeding egg 1 and the positioning detection of key parts of the breeding egg 1 (embryo 109, blood vessel 110, air chamber 111 and allantois 108, etc.), such as Figure 1 As shown, the method for positioning and detecting key parts of breeding eggs of the present invention is as follows: First, build the egg image acquisition device, such as Figure 2As shown, the breeding egg image acquisition device comprises an image collector 2, a light source 3, an egg tray 4, an egg candling device 5, a rotating bracket 6, a shielding ring 7 and a dark box. The image collector 2, the egg tray 4, the egg candling device 5, the rotating bracket 6 and the shielding ring 7 are all located in the dark box. The egg tray 4 is horizontally installed at the center bottom of the rotating bracket 6. The breeding egg 1 is vertically placed on the egg tray 4 with the blunt end facing upward. The egg candling device 5 is installed at the center top of the rotating bracket 6 and the bottom end is located directly above the breeding egg 1. The top end of the breeding egg 1 and the bottom end of the egg candling device 5 are sealed by the shielding ring 7 that isolates light. The light source 3 is installed inside the egg illuminator 5 and the light source 3 is vertically downward toward the egg 1 directly below, and the image collector 2 is installed on the side of the rotating bracket 6 and the lens is horizontally toward the egg 1; when the egg image acquisition device acquires images, the egg 1 is irradiated by the light source 3, the bracket 6 is rotated, and the rotating bracket 6 is controlled to drive the image collector 2 to rotate 360° around its own vertical central axis, and after setting the image acquisition parameters, the image collector 2 acquires the internal transmission image of the egg 1 in all directions, and then transmits it to the computer for storage. The image collector 2 can use a complementary metal oxide semiconductor CMOS (Complementary Metal-Oxide Semiconductor) industrial camera and industrial lens; the light source 3 can use an LED cold light source. Before acquisition, a 0.1% chlorhexidine solution is also required to disinfect the egg 1; when acquiring, the egg 1 is kept in a fixed position with the blunt end facing upward, and after the light source 3 is turned on, the focal length, aperture, size of the object to be measured, measurement accuracy, working distance and other parameters are adjusted to acquire the perspective information image of the egg 1 through the camera. The shielding ring 7 of the breeding egg image acquisition device is used to seal the light of the egg candling device to ensure that the light source is completely irradiated on the breeding eggs 1.

[0025] In the present invention, the focal length of the lens is determined according to the size of the conventional egg and the range of the light source. The horizontal diameter of the middle of the egg is 40-60 mm, the field of view range is 90 mm at most, the camera lens is 100 mm away from the blunt end of the egg and the middle of the egg during image acquisition, the shooting object distance D is 100 mm, and the target surface size of the camera sensor is 1.1'' (12×12 mm), so the focal length of the lens should be: f=v×D÷V=12×100÷90=13.33 mm, wherein f is the focal length of the lens, v is the longitudinal size of the camera chip, and V is the lateral size of the field of view area. In order to meet the field of view requirements, the focal length of the lens is selected to be 12 mm. The light source assembly parameters involved in the present invention are: light source: LED lamp, lamp mouth temperature: ≤10°C (under the condition of ambient temperature of 20°C), cold light source, no mercury, no ultraviolet rays and infrared rays, no heat radiation, energy saving and environmental protection, vibration resistance, impact resistance, long service life; egg penetrating power is strong, and the development of the egg 1 can be checked more clearly, and the development of the embryo 109 can be grasped. The egg illuminator 5 is designed to be cylindrical and is located between the egg 1 and the light source 3, so that the light from the light source 3 can reach the egg 1 through the channel of the egg illuminator 5, thereby reducing the scattering of the incident light on the egg 1 and the reflection of the light from other surfaces. The egg 1 is of different sizes, and there is a compressible rubber ring at the bottom of the spacer, which is closer to the eggshell surface to prevent light leakage and avoid damage. The light source is adjusted to 5-8 cm on both sides of the egg 1 to ensure that the light distribution in the egg 1 area is uniform. The egg tray 4 is provided with a circular egg hole with a diameter extending downward from 4 cm to 2 cm. The overall size of the dark box is 50 cm × 50 cm × 130 cm. It is built with black aluminum profiles and equipped with black experimental shading cloth to reduce reflections. In addition, the top of the dark box is equipped with a cooling fan to cool the camera. A piece of black velvet is also laid at the bottom of the egg tray 4 to reduce the possible impact of the background on the imaging system. The chicken embryo egg is kept in place with the blunt end facing up. After the light source 3 is turned on, the focal length, aperture, size of the object to be measured, measurement accuracy, working distance and other parameters are adjusted to obtain the perspective information image of the breeding egg 1 through the camera.

[0026] The collected breeding eggs 1 in the present invention have different categories, including infertile eggs 101, dead sperm eggs 102 and fertilized eggs 103; the breeding egg physiological state refers to the characteristics of the egg itself or the characteristics of the embryo in the egg, including unfertilized eggs or eggs containing dead, undeveloped or immature embryos. Figure 3As shown, the light source 3 shines from the blunt part of the egg 1 from top to bottom. The first time is on the fifth or sixth day after the start of incubation. The egg embryo 109 develops normally, and the blood vessels 110 are radially distributed. When the egg is illuminated, the darker the egg is, the brighter the red color is, which is a normal fertilized egg 103. If the blood vessels 110 part does not occupy four-fifths of the egg surface and the eye spot is not visible, it is determined to be a weak sperm egg. If the egg is turbid, has blood circles, and blood clots, it is a dead sperm egg 102. When the egg is illuminated, the brightness is high and only the shadow of the egg yolk color is seen, it is determined to be an unfertilized egg 101. The developmental status of the egg refers to the degree of development of the biological, biochemical and physical characteristics of the embryo in the healthy state of the egg 1. Figure 3 As shown, on the fourth day of incubation of the egg 1, the yolk sac blood vessels 110 surround nearly 1 / 3 of the yolk. The yolk is not easy to rotate when the egg is illuminated. The chicken embryo and the yolk sac blood vessels 110 are spider-like. The allantois 108 is a very small blister. The eyes begin to deposit melanin, which is the first development egg 104. On the sixth day of incubation of the egg 1, the allantois 108 grows rapidly, and the blood vessel 110 system of the allantois 108 develops rapidly. Two small round balls of the head and the trunk can be seen when the egg is illuminated, which is the second development egg 105. On the eighth day of incubation, the allantois 108 almost surrounds the yolk sac, the appearance of the chicken embryo develops gradually, and the amniotic fluid and allantoic fluid increase rapidly, which is the third development egg 106. On the tenth day of incubation, the allantois 108 develops rapidly from the back of the embryo 109 to the lower end, gradually surrounding the egg white, and the blood vessels 110 are all over the egg surface. The chicken embryo is located close to the air chamber 111, which is the fourth development egg 107. Different detection technologies are implemented according to the characteristics of the development status of the eggs and application requirements.

[0027] Then, the deep learning classification model and semantic segmentation model were trained. 800 eggs 1 on the 7th day of incubation were collected, and the transmission images of the eggs 1 were obtained using the egg image acquisition device (resolution 2048×2048, LED backlight wavelength 550±10nm). The image preprocessing first performed grayscale, denoising: Gaussian filtering (convolution kernel size is 3×3, standard deviation σ=1.5) and contrast enhancement: CLAHE (Contrast Limited Adaptive Histogram Equalization) histogram equalization (contrast limit = 2.0, block size = 8×8). Labelme software was used to add category labels fertilized eggs (category 1) and unfertilized eggs (category 0) to the transmission images inside the eggs 1, and the images were divided into training sets (640 images) and test sets (160 images) to train the deep learning classification model and semantic segmentation model.

[0028] During training, a transmission image of the inside of the egg 1 is collected using a breeding egg image acquisition device and input into the deep learning classification model. After processing, the transmission image of the fertilized egg 103 after discrimination is obtained, and then the next step of positioning and detection is performed. The deep learning classification model specifically adopts the ResNet-18 classification model, including an input layer (512×512 grayscale image), a convolution layer (7×7 convolution kernel, step size 2, output channel 64), a residual module (4 groups of residual blocks, each group contains 2 3×3 convolution layers), global average pooling, a classification layer (the fully connected layer outputs 2 neurons, corresponding to the fertilization / unfertilization probability respectively), and the activation function adopts the Softmax classifier. The classification result accuracy of the output image of the training set is 97.2%, and the accuracy of the validation set is 95.3%. The transmission image of the fertilized egg 103 after discrimination is saved as a 512×512 grayscale image for subsequent key part segmentation.

[0029] The transmission images of the fertilized eggs 103 after discrimination need to be screened, and the transmission images of the fertilized eggs 103 with embryos 109, blood vessels 110 and air chambers 111 are screened out, and the regions of the embryos 109, blood vessels 110 and air chambers 111 in each screened image are annotated, so as to be input into the semantic segmentation model for training. In specific implementation, 600 transmission image data sets of the fertilized eggs 103 after discrimination (at this time, the embryonic vascular network has been formed and the air chamber boundary is clear) are divided into training sets, verification sets and test sets according to 7:2:1 for training until the semantic segmentation model converges. When performing key part segmentation, the present invention combines the semantic segmentation algorithm to detect the development status of the breeding eggs, and the process includes loading images, preprocessing images, selecting segmentation algorithms, initializing algorithm parameters, iterative segmentation processes, outputting segmentation results until the stop condition is met, post-processing segmentation results, and displaying segmentation results. In the specific implementation, a lightweight neural network that can be used for semantic segmentation on mobile devices is selected. The available deep learning-based semantic segmentation models include the semantic segmentation model CGnet (Context Guided), the fully convolutional network FCN (Fully Convolutional Networks), the convolutional neural network SegNet (Semantic Segmentation), the convolutional neural network U-Net (U-shaped Network) and the deep convolutional neural network PSPNet (Pyramid Scene Parsing Network).

[0030] The present invention takes the semantic segmentation model CGnet as an example to identify and segment the image of the egg 1. The main process is as follows: Figure 4As shown in the figure, the semantic segmentation model CGnet introduces the context-guided block CG, which consists of a local feature extractor floc, a surrounding context extractor fsur, a joint feature extractor fjoi and a global context extractor fglo. The context-guided block CG can learn the joint features of local features and surrounding contexts, and finally introduces global context features to further improve the learning of joint features. The semantic segmentation CGNet has the ability to aggregate context information from bottom to top, reduce the number of parameters, and improve segmentation accuracy. The results show that the recognition accuracy of the key parts of the egg 1 is as follows: embryo 109: 96.8%, air chamber 111: 93.2%, main blood vessel 110: 90.5%; Intersection over Union (IoU): embryo 109: 73.3%; air chamber 111: 67.6%; main blood vessel 110: 74.1%.

[0031] After the training is completed, a breeding egg physiological state discrimination model and a breeding egg key part segmentation model are obtained, and then a breeding egg image acquisition device is used to collect a transmission image of the inside of the breeding egg 1 to be located and detected, and the transmission image of the breeding egg 1 to be located and detected is input into the breeding egg physiological state discrimination model, and after processing, a transmission image of the fertilized egg 103 after discrimination is obtained; the transmission image of the fertilized egg (103 after discrimination is input into the breeding egg key part segmentation model, and after processing, the key parts of the breeding egg 1 to be located and detected are identified, thereby obtaining a transmission image of the fertilized egg 103 after segmentation; the transmission image of the fertilized egg 103 after segmentation is located using an image processing method to extract the optimal coordinate point of the allantoic fluid, firstly, an image coordinate system is established on the transmission image of the fertilized egg 103 after segmentation in the horizontal direction and the vertical direction, thereby obtaining the contour and position coordinates of the embryo 109, the blood vessel 110 and the air chamber 111, and in specific implementation, the obtained breeding egg state image is selected, and the image contains The segmented areas of the embryo 109, the air chamber 111 and the blood vessel 110, and the contours of the embryo 109, the air chamber 111 and the blood vessel 110 in the image are extracted, and the following steps are included: graying the original image, intercepting the mask area of ​​the grayscale image, binarizing the mask area, and performing binary image operations. The edges of the embryo, the air chamber and the blood vessel area are detected using the Canny edge detection algorithm. The findContours function in OpenCV is used to extract the contours of the embryo 109, the air chamber 111 and the blood vessel 110 area, and obtain the contour coordinate set, which represents the boundaries of the embryo 109, the air chamber 111 and the blood vessel 110 area. The boundingRect function is used to obtain the minimum circumscribed rectangle of each contour, and the coordinate information of the embryo 109, the air chamber 111 and the blood vessel 110 area is obtained. The coordinate output of the 7th day breeding egg 1 measured by the present invention is: embryo center: (320±5,280±3 pixels), air chamber centroid: (150±2, 480±4 pixels), main vessel endpoint set: [(215,310), (228,305),..., (392,287)].

[0032] Then the outline of the embryo 109 is expanded outward at equal distances, with the center of the embryo as the origin, and the radius is dynamically expanded (formula: radius after expansion r′=r×(1+10 / 7), 7 is the number of incubation days), and the initial radius r is the radius of the maximum circumscribed circle of the embryo outline. After obtaining the embryo expansion area 112, the embryo expansion area 112 and the regional outline of the blood vessel 110 are compared, and the scanning line algorithm is used to calculate the overlap rate between the expansion area 112 and the blood vessel 110. When the overlap rate is >10%, coordinate elimination is triggered, and all coordinates in the repeated area of ​​the outline are set to 0, indicating that these coordinates are unavailable. After eliminating the repeated coordinate points, the remaining valid coordinate points in the expansion area 112 account for 87.7%. Input the contour coordinate set of the air chamber 111, traverse the remaining coordinate points in the contour of the embryo extension area 112 and the coordinates in the contour of the air chamber 111 area, calculate the Euclidean distance, and use the tree-shaped KDTree (k-dimensional) data structure to accelerate the nearest neighbor search to obtain the shortest Euclidean distance, so as to locate the coordinate point in the contour of the embryo extension area 112 as the optimal coordinate point for allantoic fluid extraction and mark it according to the shortest Euclidean distance. Output result: The coordinates of the allantoic fluid extraction point are (285,185) ± 3 pixels, and the positioning repeatability error is verified by 20 repeated experiments to be <2.1%.

[0033] Then, the allantoic fluid is extracted at the optimal coordinate point, and the extracted allantoic fluid of the egg 1 is spectrally scanned to obtain spectral data for testing the egg 1. The information of the egg 1 is detected by using spectral non-destructive testing technology or molecular biology minimally invasive testing technology to perform subsequent sex identification, development evaluation and quality testing of the egg 1. Non-destructive testing can be performed on the key parts of the egg 1, such as the allantoic 108, embryo 109, and blood vessels 110, based on spectral technology, and a non-destructive testing model can be constructed in combination with machine learning for sex identification, egg development evaluation, egg quality testing, etc. Spectral technology includes visible near-infrared spectroscopy, Fourier transform near-infrared spectroscopy, Raman spectroscopy, infrared spectroscopy, and terahertz spectroscopy.

[0034] After locating the allantoic 108 of the breeding egg 1, the present invention extracts the allantoic fluid through an extractor, and evaluates the allantoic fluid extraction rate and the survival rate of the breeding egg 1 at different incubation days to ensure the reliability of the method of the present invention. The allantoic fluid is detected based on molecular biological methods to achieve gender identification, breeding egg development evaluation, breeding egg quality detection, etc. Molecular biological methods include real-time fluorescence quantitative PCR (Quantitative Real-time polymerase chain reaction), immunoassay, liquid chromatography LC (Liquid Chromatograph) and mass spectrometry MS (Liquid Chromatograph) analysis, etc. The allantoic fluid extraction rate is defined as the ratio of successful extraction to the total number of attempts, combined with the hatching rate of the breeding egg 1, as a dual indicator to evaluate the feasibility of the extraction method. Figure 5 As shown, the test results show that the allantoic fluid extraction method of the present invention has been significantly improved compared with the currently used air chamber extraction method. Specifically, on the 9th day of chicken embryo incubation, the allantoic fluid extraction rate of 280 eggs 1 was greater than 90%, and the hatching rate of egg 1 was greater than 92% (compared with 200 eggs in the untreated hatching group: hatching rate 96%; 200 eggs in the traditional extraction method hatching group: the allantoic fluid extraction rate of egg 1 was less than 80%, and the hatching rate of egg 1 was about 88%). The evaluation results of the double standard strongly prove the effectiveness of the method of the present invention.

[0035] Ultimately, the allantoic fluid can be tested based on molecular biology methods, which can be applied to multimodal hatching applications such as egg sex identification, developmental assessment, and quality inspection. Compared with a single detection method, this system combines the physical characteristics of image positioning with the biochemical characteristics of allantoic fluid, providing dual verification for early evaluation of egg hatching.

[0036] The present invention also constructs a breeding egg key part positioning and detection system, including a data acquisition processing unit and a key part positioning and detection unit. The data acquisition processing unit is used to obtain a transmission image inside the breeding egg 1, and obtain a transmission image of the fertilized egg 103 after being judged after being processed by a breeding egg physiological state discrimination model and a breeding egg key part segmentation model; the key part positioning and detection unit is used to process the transmission image of the fertilized egg 103 by a breeding egg key part segmentation model to obtain a segmented transmission image of the fertilized egg 103, and then use an image processing method to locate the optimal coordinate point for allantoic fluid extraction, and extract the allantoic fluid according to the optimal coordinate point for subsequent detection.

[0037] It should be understood by those skilled in the art that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages. The present application is described according to the flowcharts of the methods, systems and computer program products of the embodiments of the present application.

[0038] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the present invention is intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0039] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of equivalent technologies of the present invention, the present application is also intended to include these modifications and variations.

Claims

1. A method for locating and detecting key parts of breeding eggs, characterized in that: include: Step 1, building a breeding egg image acquisition device, collecting transmission images of the inside of a number of breeding eggs (1) of different days, and respectively obtaining a breeding egg physiological state discrimination model and a breeding egg key part segmentation model after training through a deep learning classification model and a semantic segmentation model; Step 2, using a breeding egg image acquisition device to acquire a transmission image of the interior of the breeding egg (1) to be located and detected, inputting the transmission image of the breeding egg (1) to be located and detected into a breeding egg physiological state discrimination model, and obtaining a transmission image of a fertilized egg (103) after discrimination after processing; Step 3, inputting the identified transmission image of the fertilized egg (103) into the breeding egg key part segmentation model, and identifying the key parts of the breeding egg (1) to be located and detected after processing, thereby obtaining the segmented transmission image of the fertilized egg (103); Step 4, using an image processing method to locate the optimal coordinate point for allantoic fluid extraction from the transmission image of the segmented fertilized egg (103), extracting the allantoic fluid at the optimal coordinate point, and detecting the breeding egg (1) through the extracted allantoic fluid.

2. The method for locating and detecting key parts of eggs according to claim 1, characterized in that: In the step 1, the egg image acquisition device comprises an image acquisition device (2), a light source (3), an egg tray (4), an egg candling device (5), a rotating bracket (6), a shielding ring (7) and a dark box. The image acquisition device (2), the egg tray (4), the egg candling device (5), the rotating bracket (6) and the shielding ring (7) are all located in the dark box. The egg tray (4) is horizontally mounted at the center bottom of the rotating bracket (6). The egg (1) is vertically placed on the egg tray (4). The egg candling device (5) is mounted at the center top of the rotating bracket (6) and its bottom end is located directly above the egg (1). The top end of the egg (1) and the bottom end of the egg candling device (5) are separated by an insulating layer. The light shielding ring (7) is sealed, the light source (3) is installed inside the egg illuminator (5) and the light source (3) is vertically downwardly directed toward the breeding egg (1) directly below, and the image collector (2) is installed on the side of the rotating bracket (6) and the lens is horizontally directed toward the breeding egg (1); when the breeding egg image acquisition device acquires an image, the breeding egg (1) is illuminated by the light source (3), the rotating bracket (6) is rotated, and the rotating bracket (6) is controlled to drive the image collector (2) to rotate 360 ​​degrees around its own vertical center axis, and the transmission image of the inside of the breeding egg (1) in all directions is acquired by the image collector (2), and then transmitted to a computer for storage.

3. The method for locating and detecting key parts of eggs according to claim 1, characterized in that: In step 1, the breeding eggs (1) are all eggs that have been incubated for 3-15 days.

4. The method for locating and detecting key parts of eggs according to claim 1, characterized in that: In the step 1, different categories of breeding eggs (1) include unfertilized eggs (101), dead sperm eggs (102) and fertilized eggs (103); after adding category labels to the transmission images inside the various breeding eggs (1), the deep learning model is trained to obtain a breeding egg physiological state discrimination model, so as to discriminate the category of the breeding eggs (1) according to the breeding egg physiological state discrimination model, obtain the transmission image of the fertilized egg (103), and then carry out the next step of positioning and detection.

5. The method for locating and detecting key parts of eggs according to claim 1, characterized in that: In the step 3, the transmission images of the identified fertilized eggs (103) are screened to select the transmission images of the fertilized eggs (103) having embryos (109), blood vessels (110) and air chambers (111). The regions of the embryos (109), blood vessels (110) and air chambers (111) in each of the screened images are labeled and then input into a semantic segmentation model for training to obtain a segmentation model of key parts of the breeding eggs.

6. The method for locating and detecting key parts of eggs according to claim 1, characterized in that: In the step 4, the transmission image of the segmented fertilized egg (103) is processed using an image processing method to obtain the optimal coordinate points for allantoic fluid extraction. First, an image coordinate system is established on the transmission image of the segmented fertilized egg (103) to obtain the contours and position coordinates of the embryo (109), blood vessels (110) and air chambers (111); the contour of the embryo (109) is expanded at equal distances to obtain the embryo expansion area (112); the regional contours of the embryo expansion area (112) and the blood vessels (110) are compared. If If there are repeated areas in the contour, all coordinate points in the repeated areas of the contour are set as repeated coordinate points, and the repeated coordinate points are removed. The remaining coordinate points in the contour of the embryonic expansion area (112) and the coordinates in the contour of the air chamber (111) are traversed, and the Euclidean distance is calculated. The tree-shaped KDTree data structure is used to accelerate the nearest neighbor search to obtain the shortest Euclidean distance, so that the coordinate points in the contour of the embryonic expansion area (112) are located according to the shortest Euclidean distance as the optimal coordinate points for allantoic fluid extraction and marked.

7. The method for locating and detecting key parts of eggs according to claim 1, characterized in that: In the step 4, the allantoic fluid extracted from the breeding eggs (1) is spectrally scanned to obtain spectral data for testing the breeding eggs (1).

8. A system for positioning and detecting key parts of hatching eggs suitable for the method according to any one of claims 1 to 7, characterized in that: include: A data acquisition processing unit, used to acquire a transmission image of the interior of a breeding egg (1), and obtain a transmission image of a fertilized egg (103) after being identified through processing using a breeding egg physiological state identification model and a breeding egg key part segmentation model; The key part positioning and detection unit is used to obtain a segmented transmission image of the fertilized egg (103) after processing the transmission image of the fertilized egg (103) through the key part segmentation model of the breeding egg, and then use the image processing method to locate the optimal coordinate point for allantoic fluid extraction, and extract the allantoic fluid according to the optimal coordinate point for subsequent detection.

9. An electronic device, characterized in that: include: A memory and a processor coupled to each other, wherein the memory stores program data, and the processor calls the program data to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having program data stored thereon, characterized in that: When the program data is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images