Method and equipment for detecting insemination of hatching eggs before hatching based on magnetic resonance

Abstract

The invention discloses a method and equipment for detecting infertility of hatching eggs before hatching based on magnetic resonance, and relates to the field of poultry breeding industry, and the method comprises the steps: transmitting the hatching eggs through a transmission module based on an assembly line, and transmitting a signal to a main control module when the passing of the hatching eggs is detected; the main control module records the position of the hatching egg, controls the transmission module to stop when the hatching egg enters an imaging area, and sends an instruction to the image acquisition module; the image acquisition module acquires nuclear magnetic images of hatching eggs to obtain high-resolution images of internal structures of the hatching eggs; the image processing and analyzing module is used for detecting the fertilization state of the hatching egg and transmitting the fertilization state detection result of the hatching egg to the main control module; the main control module controls the transmission module to continue transmission, when the hatching eggs enter the sorting area, the transmission module is stopped again, and a sorting instruction is sent to the sorting module according to the detection result; and the sorting module classifies the hatching eggs and places the hatching eggs to corresponding positions according to a classification result. The invention solves the limitation of the existing detection method.

Description

Technical Field

[0001] The invention relates to the field of poultry farming, and in particular to a method and equipment for detecting sperm-free breeding eggs before hatching based on magnetic resonance. Background Art

[0002] For poultry farming companies, the incubation of breeding eggs is a key link in the entire industry. Due to the limitations of the production environment and poultry breeds, under natural mating conditions, even for high-quality breeds such as Beijing ducks, the fertilization rate is only about 90%; and for other breeds such as Nandan Yao chickens, the average fertilization rate is only 92.59%. This means that about 10% of the eggs are unfertilized during the incubation process. These unfertilized eggs not only take up valuable incubation space and resources, but may also deteriorate and rot during the incubation process, producing odor and bacterial contamination, which will have an adverse effect on the fertilized eggs. Not only that, even after the incubation is over, these unfertilized eggs can no longer be used as food and will eventually be wasted.

[0003] In order to solve the above problems, researchers at home and abroad have developed a variety of methods and devices for identifying unfertilized eggs, but they all have certain limitations. CN202210441509.4 and CN202010547278.6 use computer vision technology to detect infertile eggs, but this method cannot accurately distinguish between fertilized eggs and infertile eggs before hatching. CN201510994891.1 uses an X-ray device to detect the activity of eggs, but it is also unable to detect the results before the eggs are incubated, and cannot meet actual production needs. Since infertile eggs deteriorate rapidly during the incubation process, even if they are identified after hatching, these eggs can no longer enter the market, which will eventually lead to a waste of egg resources. CN201610446950.6 judges the fertilization of eggs based on the amount of carbon dioxide released, but it is easily disturbed by external environmental factors and has low stability. CN202010704135.1 and CN202010387554.7 use fiber spectroscopy technology and convolutional neural networks to detect the fertilization information of breeding duck eggs, but the device only has a detection part and lacks a subsequent automatic sorting process, so it cannot be applied to industrial production. CN202311039209.4 uses an optical light box device to pick out some fertilized eggs at one time, but this method is not suitable for large-scale breeding applications. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method and equipment for detecting sperm-free eggs before incubation based on magnetic resonance, which solves the limitations of the prior art methods.

[0005] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is: a method for detecting sperm-free eggs before incubation based on magnetic resonance, comprising the following steps: S1: The eggs are transmitted using a pipeline-based transmission module. When the infrared sensor detects that the eggs have passed by, a signal is sent to the main control module. S2: Using the main control module to record the position of the eggs, when the eggs enter the imaging area, controlling the pipeline-based transmission module to stop, and sending instructions to the image acquisition module based on the magnetic resonance device; S3: using an image acquisition module based on a magnetic resonance device to acquire nuclear magnetic images of the breeding eggs, and obtaining a high-resolution image of the internal structure of the breeding eggs; S4: using the image processing and analysis module to detect the fertilization status of the breeding eggs according to the high-resolution image of the internal structure of the breeding eggs, and transmitting the detection result of the fertilization status of the breeding eggs to the main control module; S5: using the main control module to control the pipeline-based transmission module to continue transmission, when the eggs enter the sorting area, the pipeline-based transmission module is stopped again, and a sorting instruction is sent to the sorting module according to the result of the fertilization status detection of the eggs; S6: using the sorting module to classify the breeding eggs according to the test results of the breeding egg fertilization status, and placing the breeding eggs in corresponding positions according to the classification results; S7: Repeat the above steps S1-S6 until all the breeding eggs are tested, and the pre-hatching breeding egg sperm-free detection based on magnetic resonance is completed.

[0006] Furthermore, the S3 adopts an improved MA-Unet network based on the UNet network to obtain a high-resolution image of the internal structure of the egg. The MA-Unet network adds a convolutional attention mechanism module CBAM after each encoder convolution layer of the UNet network, and adds a hole space pyramid pooling module ASPP after the feature map with a channel number of 512 at the end of the encoder. The feature maps of each level are connected in series in the decoder.

[0007] Furthermore, the S4 uses an Egg-EfficientNet network improved based on EfficientNet to detect the fertilization status of eggs, and the Egg-EfficientNet network includes a 3×3 convolution block, a Fused-MBConv1 module, a Fused-MBConv4 module, a Fused-MBConv4 module, an MBConv4 module, an MBConv6 module, an MBConv6 module and a 1×1 convolution block, a pooling layer and a fully connected layer connected in sequence; The Hard-Swish activation function is added to the Fused-MBConv1 module, and the formula is:

[0008] in, is the Hard-Swish activation function, is the ReLU activation function, is a variable; The SE attention mechanism is added to the MBConv4 module and the MBConv6 module.

[0009] Furthermore, the fertilization status of the eggs in S4 includes fertilized eggs, infertile eggs and abnormal eggs, which are classified according to the confidence score of the fertilization accuracy of each egg, including the confidence score of the fertilized eggs. , confidence level of infertile eggs and confidence level of abnormal eggs , the confidence level of the fertilized egg , confidence level of infertile eggs and confidence level of abnormal eggs All of them are within the range of 0 to 1 and the sum of the three confidence levels is 1, and the category with the maximum confidence level is taken as the result of the egg fertilization status detection.

[0010] The present invention also adopts a technical solution: a pre-hatching egg spermless detection device based on magnetic resonance, comprising a conveyor belt, an egg tray, an infrared sensor, a nuclear magnetic resonance instrument, a suction cup, a suction cup swing arm and an egg collection table, wherein the central area inside the nuclear magnetic resonance instrument probe is an imaging area; The conveyor belt is partially mounted on a bracket, and the bracket passes through the imaging area and the bottom opening of the nuclear magnetic resonance apparatus. The egg tray is fixed on the conveyor belt, the infrared sensor is fixed on the bracket on the left side of the imaging area, and the suction cup is fixed on the end of the suction cup swing arm. The suction cup swing arm is located directly above the bracket on the right side of the imaging area. The position of the egg tray on the right side of the imaging area is the sorting area.

[0011] Furthermore, the conveyor belt is made of wear-resistant PVC material, and adjustable edge guide devices are arranged on both sides. The conveyor belt is driven by a stepper motor and cooperates with an infrared sensor to realize egg position feedback.

[0012] Furthermore, the suction cup swing arm is driven by a pneumatic push rod.

[0013] The beneficial effects of the present invention are: (1) Magnetic resonance image accelerated acquisition method based on MA-Unet. Applying the image reconstruction algorithm to image acquisition not only reduces the image acquisition time of the magnetic resonance device, but also ensures the acquisition of clear images, greatly improving the performance of the system. This non-destructive imaging method can obtain high-resolution image data of the internal structure of each egg and comprehensively analyze the fertilization status of the egg, which is more accurate and reliable than traditional appearance inspection. This innovative solution brings new possibilities for internal quality monitoring of eggs and deserves further exploration and promotion among a wider range of industrial application manufacturers.

[0014] (2) Egg-EfficientNet-based method for detecting fertilization information of pre-hatched eggs. The system uses advanced deep learning algorithms to accurately identify the fertilization status of eggs. Through a large amount of training data, the algorithm can give a confidence score for the fertilization accuracy of each egg, providing a basis for sorting. The parallel computing capability of the GPU server cluster ensures that image analysis can keep up with the speed of the pipeline in real time.

[0015] (3) Accurate and efficient sorting method. The intelligent pneumatic suction cup swing arm sorting mechanism can quickly and accurately sort the eggs. The sorting process is automatically completed under PLC control without manual intervention, which greatly improves work efficiency. The reliability and repeated positioning accuracy of the high-speed sorting mechanism ensure the stability of the sorting quality.

[0016] (4) System integration and intelligent control. Each subsystem uses industrial Ethernet communication to achieve seamless integration and real-time data sharing. The central control system provides functions such as human-machine interaction interface, remote monitoring, and fault diagnosis, which greatly improves the intelligence level of the system. Advanced automation technology significantly reduces dependence on manual operation and reduces labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The present invention is a flow chart of a method for detecting sperm-free pre-hatching eggs based on magnetic resonance.

[0018] Figure 2 This is a network structure diagram of the image reconstruction algorithm of the present invention.

[0019] Figure 3 This is a network structure diagram of the fertilization detection algorithm of the present invention.

[0020] Figure 4 This is a diagram of the key technical architecture of the present invention.

[0021] Figure 5 The present invention is a schematic diagram of the mechanical structure of a magnetic resonance-based pre-hatching egg sperm-free detection device.

[0022] Among them: 1. Conveyor belt; 2. Egg tray; 3. Infrared sensor; 4. Nuclear magnetic resonance instrument; 5. Suction cup; 6. Suction cup swing arm; 7. Egg collection table; 8. Imaging area; 9. Sorting area. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0024] Embodiment 1, as Figure 1 As shown, a method for detecting sperm-free pre-hatched eggs based on magnetic resonance comprises the following steps: S1: The eggs are transmitted using a pipeline-based transmission module. When the infrared sensor detects that the eggs have passed by, a signal is sent to the main control module. The transmission module based on the assembly line: a single-layer conveyor belt 1 with modular design can simultaneously transmit three eggs per row. The conveyor belt 1 is made of wear-resistant PVC material with a smooth surface to prevent scratches or damage to the eggs during transmission. Adjustable edge guides are set on both sides of the conveyor belt 1 to ensure that the eggs are always kept in the correct conveying position. The conveyor belt 1 is driven by a high-precision stepper motor and cooperates with a photoelectric encoder to achieve position feedback. The PLC accurately controls the transmission speed. An infrared sensor 3 is set at the starting point of the conveyor belt 1 to detect the entry of the eggs and provide real-time feedback for the coordinated work of downstream equipment.

[0025] S2: Using the main control module to record the position of the eggs, when the eggs enter the imaging area, controlling the pipeline-based transmission module to stop, and sending instructions to the image acquisition module based on the magnetic resonance device; The main control module is deployed on an independent high-performance industrial computer. The overall system uses industrial Ethernet technology for communication. Each submodule exchanges and controls data through standard industrial protocols, integrating functions such as human-machine interface, data storage, and fault diagnosis. The human-machine interface uses a high-definition display screen to display the operating status, statistical data, and alarm information of each module in real time. The main control module also provides remote monitoring and management functions. Technicians can check the equipment status at any time through the network for centralized maintenance and optimization. Finally, the main control module also has a fault self-diagnosis function, which can quickly locate and isolate the fault point to ensure the high reliability of the overall system.

[0026] S3: using an image acquisition module based on a magnetic resonance device to acquire nuclear magnetic images of the breeding eggs, and obtaining a high-resolution image of the internal structure of the breeding eggs; Image acquisition module based on magnetic resonance equipment: Combined with the innovative method of building an image reconstruction algorithm based on MA-Unet to speed up the magnetic resonance acquisition speed, significant optimization has been achieved in image acquisition and processing. This algorithm not only reduces the image acquisition time of the magnetic resonance equipment, but also ensures that clear images are obtained, greatly improving the performance of the system. Based on the above algorithm, a 0.3T low-field nuclear magnetic resonance instrument is equipped, and complex image reconstruction calculations are performed, and finally high-resolution image data of the internal structure of each egg can be obtained. These image data contain detailed information on the fertilization status of the egg. In addition, the image acquisition module, transmission module and sorting module are synchronized in real time to ensure that the image corresponds to the position of the egg. Industrial-grade Ethernet communication is used to transmit high-definition image data to the subsequent image processing server in real time.

[0027] S4: using the image processing and analysis module to detect the fertilization status of the breeding eggs according to the high-resolution image of the internal structure of the breeding eggs, and transmitting the detection result of the fertilization status of the breeding eggs to the main control module; Image processing and analysis module: Using the Egg-EfficientNet network, the model trained with a large number of egg samples can accurately identify the fertilization status of the eggs. The algorithm can not only determine the fertilization status, but also give a confidence score for the accuracy of each egg, and distinguish between fertilized eggs, infertile eggs, and abnormal eggs. Image analysis uses a high-performance GPU server cluster for distributed parallel computing, which can process high-definition image data in real time with a delay of less than 100ms. The analysis results are transmitted to the main control system in real time via Ethernet communication, providing an accurate basis for downstream precision sorting.

[0028] S5: using the main control module to control the pipeline-based transmission module to continue transmission, when the eggs enter the sorting area, the pipeline-based transmission module is stopped again, and a sorting instruction is sent to the sorting module according to the result of the fertilization status detection of the eggs; S6: using the sorting module to classify the breeding eggs according to the test results of the breeding egg fertilization status, and placing the breeding eggs in corresponding positions according to the classification results; Sorting module: The suction cup swing arm sorting mechanism driven by pneumatic push rod can quickly sort eggs without reducing the transmission speed. The sorting module can sort eggs into three categories: normal fertilized eggs, infertile eggs and abnormal eggs. The sorting module accurately sorts the eggs to the corresponding positions according to the control instructions of the main control module. The sorting process is automatically carried out under the real-time coordinated control of PLC without manual intervention.

[0029] S7: Repeat the above steps S1-S6 until all the breeding eggs are tested, and the pre-hatching breeding egg sperm-free detection based on magnetic resonance is completed.

[0030] In one embodiment of the present invention, when the equipment is turned on, the nuclear magnetic resonance instrument 4 enters the startup and self-test state, and the infrared sensor 3 and the sorting module are turned on. Subsequently, the breeding eggs are placed in the conveyor belt 1, and the entire device continues to operate. The infrared sensor 3 determines whether the breeding eggs pass through. If they pass through, a signal is sent to the main control module, and the main control module records the position of the breeding eggs. When the breeding eggs enter the imaging area 8, the conveyor belt 1 is controlled to stop, and a command is sent to the nuclear magnetic resonance instrument 4 to start collecting nuclear magnetic images of the breeding eggs; if they do not pass through, the conveyor belt 1 continues to run until a breeding egg passes through.

[0031] During the acquisition process, the collected images will be reconstructed by the image reconstruction algorithm to obtain a high-resolution image of the internal structure of the eggs, and then handed over to the fertilization status detection algorithm for a comprehensive analysis of the fertilization status of the eggs.

[0032] In this embodiment, in order to speed up the magnetic resonance imaging, undersampling is performed during the image acquisition process to reduce the sampling time, but this results in a decrease in image quality and artifacts, so the MA-Unet network is required for image reconstruction, and finally a high-resolution image of the internal structure of the egg is obtained. The specific operation of undersampling is to select only a portion of the data for sampling when controlling the nuclear magnetic resonance instrument to acquire K-space (frequency domain) data.

[0033] In the S3, an improved MA-Unet network based on the UNet network is used to obtain a high-resolution image of the internal structure of the egg. The MA-Unet network adds a convolutional attention mechanism module CBAM after each encoder convolution layer of the UNet network, and adds a hole space pyramid pooling module ASPP after the feature map with a channel number of 512 at the end of the encoder. The feature maps of each level are connected in series in the decoder.

[0034] The image reconstruction algorithm used in the present invention is constructed based on the MA-Unet network improved by the UNet network. The network structure is as follows: Figure 2 As shown in the figure, compared with the UNet network, MA-UNet adds a convolutional attention mechanism module (CBAM). To address the problem of some invalid semantic information carried by UNet during jump connections, a CBAM module is introduced after each encoding layer convolution layer. CBAM consists of two key parts, namely the channel attention mechanism and the spatial attention mechanism. The spatial attention mechanism focuses on enhancing the spatial information in the upsampled feature map, which can better capture the spatial structure of the embryonic disc in the nuclear magnetic resonance image. Secondly, the atrous spatial pyramid pooling module (ASPP) is added. In order to enhance the ability of UNet in capturing multi-scale information, an improved ASPP module is integrated after the feature map with 512 channels at the end of the encoder. This module can capture features at different scales, thereby improving the accuracy of image reconstruction and the ability to restore details. For the decoder part, in order to better handle the multi-scale problem of targets in the image, the feature maps of each level are connected in series to ensure that in the final stage of feature extraction, the feature map can comprehensively reflect the feature information of targets of different scales. Finally, the mean square error loss MSE is used as the loss function. The calculation formula of MSE is:

[0035] in, For the The true value of the samples, For the The predicted value of samples, is the number of samples in the dataset.

[0036] In S4, an Egg-EfficientNet network improved based on EfficientNet is used to detect the fertilization status of breeding eggs, and the Egg-EfficientNet network includes a 3×3 convolution block, a Fused-MBConv1 module, a Fused-MBConv4 module, a Fused-MBConv4 module, an MBConv4 module, an MBConv6 module, an MBConv6 module and a 1×1 convolution block, a pooling layer and a fully connected layer connected in sequence; The Hard-Swish activation function is added to the Fused-MBConv1 module, and the formula is:

[0037] in, is the Hard-Swish activation function, is the ReLU activation function, is a variable; The SE attention mechanism is added to the MBConv4 module and the MBConv6 module.

[0038] The present invention improves the structure of EfficientNet and names it Egg-EfficientNet. The network structure is as follows: Figure 3As shown in the figure. Since the embryonic disc of the breeding egg is small, the use of ordinary convolution will result in the location information of the embryonic disc not being paid attention to, thus losing some information, resulting in insufficient detection accuracy. The network model of the Egg-EfficientNet structure introduces the Fused-MBConv module, replacing the Expansion Conv 1×1 and DepthwiseConv 3×3 in the original MBConv main branch with an ordinary Conv 3×3. The replaced network model has fewer parameters, faster training, and can pay more attention to the embryonic disc of the breeding egg. At the same time, a progressive learning strategy is adopted. The regularization method is dynamically adjusted according to the size of the training image to improve the training speed and accuracy. In addition, the SE attention mechanism is introduced to help the model locate and identify the information of the embryonic disc area of ​​the breeding egg more comprehensively and accurately, effectively suppress the influence of background information on the detection results, make the model classification more accurate, and solve the problem of low classification accuracy due to insufficient feature information. In order to further improve the classification accuracy, the Hard-Swish activation function is also introduced. This activation function can not only speed up the calculation speed of Egg-EfficientNet, but also effectively improve the classification accuracy. At the same time, the improved model is equipped with a Dropout layer and a normalization layer, which can better suppress overfitting, speed up model convergence, improve model robustness, and significantly improve classification accuracy. Finally, in order to accelerate model reasoning, pruning, knowledge distillation, and TensorRT technology are combined to compress the model, increasing the processing speed by more than 3 times.

[0039] In the specific implementation process, in the early stage of work, more than 2,000 nuclear magnetic resonance images of breeding eggs can be collected to build a data set for training the network. After manual classification, the breeding eggs are divided into infertile eggs, fertilized eggs and abnormal eggs. Finally, the error calculated by the loss function of the model is back-propagated to update the network parameters.

[0040] More specifically, the data set is divided into training set, test set and validation set in the ratio of 8:1:1. After the eggs are classified by the Egg-EfficientNet network, the recognition results are fed back to the main control module, which then controls the sorting module for classification.

[0041] The fertilization status of the eggs in S4 includes fertilized eggs, infertile eggs and abnormal eggs, which are classified by the confidence score of the fertilization accuracy of each egg. For each egg, the model will output three confidence scores, including the confidence score of the fertilized egg. , confidence level of infertile eggs and confidence level of abnormal eggs , the confidence level of the fertilized egg , confidence level of infertile eggs and confidence level of abnormal eggs If all three confidence levels are within the range of 0 to 1 and the sum of the three confidence levels is 1, the category with the highest confidence level is taken as the result of the fertilization status test of the egg. If no clear category wins, the egg may be classified as abnormal or require further testing.

[0042] The technology stack for intelligent detection of infertile eggs before hatching based on nuclear magnetic resonance image information of hatching eggs is as follows: Figure 4 As shown, the main control module sends instructions to complete the entire detection and sorting process. Figure 1 As shown. When the breeding eggs are placed in the conveyor belt 1 in a good posture, the first device that passes through before entering the magnetic inspection is the infrared sensor 3, which is used to sense the breeding eggs, and the main control module records the position of the breeding eggs. After detecting that the breeding eggs have passed through the imaging area 8, the conveyor belt 1 is stopped and the nuclear magnetic image of the breeding eggs is collected. After the test, the fertilization information of the breeding eggs is obtained and uploaded to the server for display, and the conveyor belt 1 is restarted. When the breeding eggs pass through the sorting area 9, the conveyor belt 1 is stopped again, and the main control module sends a sorting instruction to the suction cup swing arm mechanism according to the fertilization information results of the breeding eggs, classifies the fertilized eggs, unfertilized eggs and abnormal eggs, and puts them in the corresponding positions in the breeding egg collection table 7.

[0043] Embodiment 2, as Figure 5 As shown, a pre-hatching egg spermless detection device based on magnetic resonance includes a conveyor belt 1, an egg tray 2, an infrared sensor 3, a nuclear magnetic resonance instrument 4, a suction cup 5, a suction cup swing arm 6 and an egg collection table 7, wherein the central area inside the probe of the nuclear magnetic resonance instrument 4 is an imaging area 8; The conveyor belt 1 is partially mounted on a bracket, and the bracket passes through the imaging area 8 and the bottom opening of the nuclear magnetic resonance apparatus 4. The egg tray 2 is fixed on the conveyor belt 1. The infrared sensor 3 is fixed on the bracket on the left side of the imaging area 8. The suction cup 5 is fixed on the end of the suction cup swing arm 6. The suction cup swing arm 6 is located directly above the bracket on the right side of the imaging area 8. The position of the egg tray 2 on the right side of the imaging area 8 is the sorting area 9.

[0044] The conveyor belt 1 is made of wear-resistant PVC material and is provided with adjustable edge guide devices on both sides. The conveyor belt 1 is driven by a stepper motor and cooperates with an infrared sensor 3 to realize egg position feedback.

[0045] The suction cup swing arm 6 is driven by a pneumatic push rod. When the eggs pass through the sorting area 9, the suction cup swing arm 6 sucks the eggs and moves them to the egg collection table 7. The suction cup swing arm 6 realizes the suction and movement of the eggs. The suction cup uses the air pressure difference to form an adsorption force to firmly adsorb the eggs on its surface. The pneumatically driven system responds quickly and can improve production efficiency. The suction cup 5 can also be accurately positioned as needed to reduce the collision and damage of the eggs during the movement. At the same time, it can handle eggs of different shapes and sizes to meet various production needs.

[0046] The above-mentioned equipment and system designed for the needs of pre-hatching sorting of breeding eggs, and the assembly line control logic designed for nuclear magnetic resonance data and equipment characteristics are designed for the present invention. However, the design concept of the present invention is not limited to the above design. The design concept of the assembly line of the present invention is based on the non-overlapping nature of the nuclear magnetic resonance data of breeding eggs, and the control logic designed according to the characteristics of the nuclear magnetic resonance data itself, without the need to interact with other information of the breeding eggs, directly outputs the pre-hatching fertilization information of the breeding eggs, and can automatically complete the detection and sorting of the breeding eggs without human intervention. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for detecting sperm-free pre-hatching eggs based on magnetic resonance, which is implemented by detecting sperm-free pre-hatching eggs based on magnetic resonance, wherein the detecting sperm-free pre-hatching eggs comprises a conveyor belt (1), an egg tray (2), an infrared sensor (3), a nuclear magnetic resonance instrument (4), a suction cup (5), a suction cup swing arm (6) and an egg collection table (7), wherein the central area inside the probe of the nuclear magnetic resonance instrument (4) is an imaging area (8); The conveyor belt (1) is partially mounted on a bracket, the bracket passes through the imaging area (8) and the bottom opening of the nuclear magnetic resonance apparatus (4), the egg tray (2) is fixed on the conveyor belt (1), the infrared sensor (3) is fixed on the bracket on the left side of the imaging area (8), the suction cup (5) is fixed on the end of the suction cup swing arm (6), the suction cup swing arm (6) is located directly above the bracket on the right side of the imaging area (8), and the position of the egg tray (2) on the right side of the imaging area (8) is a sorting area (9); The conveyor belt (1) is made of wear-resistant PVC material and is provided with adjustable edge guide devices on both sides. The conveyor belt (1) is driven by a stepper motor and cooperates with an infrared sensor (3) to realize egg position feedback. The suction cup swing arm (6) is driven by a pneumatic push rod; It is characterized in that The method for detecting sperm-free eggs before hatching comprises the following steps: S1: The eggs are transmitted using a pipeline-based transmission module. When the infrared sensor detects that the eggs have passed by, a signal is sent to the main control module. S2: Using the main control module to record the position of the eggs, when the eggs enter the imaging area, controlling the pipeline-based transmission module to stop, and sending instructions to the image acquisition module based on the magnetic resonance device; S3: using an image acquisition module based on a magnetic resonance device to acquire nuclear magnetic images of the breeding eggs, and obtaining a high-resolution image of the internal structure of the breeding eggs; S4: using the image processing and analysis module to detect the fertilization status of the breeding eggs based on the high-resolution image of the internal structure of the breeding eggs, and transmitting the detection result of the fertilization status of the breeding eggs to the main control module; S5: using the main control module to control the pipeline-based transmission module to continue transmission, when the eggs enter the sorting area, the pipeline-based transmission module is stopped again, and a sorting instruction is sent to the sorting module according to the result of the fertilization status detection of the eggs; S6: using the sorting module to classify the breeding eggs according to the test results of the breeding egg fertilization status, and placing the breeding eggs in corresponding positions according to the classification results; S7: repeating the above steps S1-S6 until all the eggs are tested, and the pre-hatching egg sperm-free test based on magnetic resonance is completed; In the S3, an improved MA-Unet network based on the UNet network is used to obtain a high-resolution image of the internal structure of the egg. The MA-Unet network adds a convolutional attention mechanism module CBAM after each encoder convolution layer of the UNet network, and adds a hole space pyramid pooling module ASPP after the feature map with a channel number of 512 at the end of the encoder. The feature maps of each level are connected in series in the decoder.

2. A method for detecting sperm-free eggs before hatching based on magnetic resonance according to claim 1, characterized in that: In S4, an Egg-EfficientNet network improved based on EfficientNet is used to detect the fertilization status of breeding eggs, and the Egg-EfficientNet network includes a 3×3 convolution block, a Fused-MBConv1 module, a Fused-MBConv4 module, a Fused-MBConv4 module, an MBConv4 module, an MBConv6 module, an MBConv6 module and a 1×1 convolution block, a pooling layer and a fully connected layer connected in sequence; The Hard-Swish activation function is added to the Fused-MBConv1 module, and the formula is: in, is the Hard-Swish activation function, is the ReLU activation function, is a variable; The SE attention mechanism is added to the MBConv4 module and the MBConv6 module.

3. A method for detecting sperm-free eggs before hatching based on magnetic resonance according to claim 1, characterized in that: The fertilization status of the eggs in S4 includes fertilized eggs, infertile eggs and abnormal eggs, which are classified according to the confidence score of the fertilization accuracy of each egg, including the confidence score of the fertilized eggs. , confidence level of infertile eggs and confidence of abnormal eggs , the confidence level of the fertilized egg , confidence level of infertile eggs and confidence of abnormal eggs All of them are within the range of 0 to 1 and the sum of the three confidence levels is 1, and the category with the maximum confidence level is taken as the result of the egg fertilization status detection.

Images