Beef cattle feeding management method and system based on machine learning

Abstract

The present invention provides a method and system for feeding and managing beef cattle based on machine learning, and relates to the technical field of data management. The method comprises: determining the growth stage; taking a first image and a second image, and weighing a first body weight and a second body weight; obtaining a feed ratio and a feed feeding amount; processing the first image and the second image by a growth morphology abnormality recognition model to obtain a growth morphology abnormality score; determining a growth abnormality score according to the growth morphology abnormality score, the first body weight and the second body weight; judging whether the feed ratio and the feed feeding amount need to be adjusted, and if adjustment is required, obtaining the adjusted feed ratio and the adjusted feed feeding amount by a feed optimization model. According to the present invention, whether the growth condition of beef cattle is abnormal can be judged by a growth morphology abnormality recognition model, and the feed ratio and the feed feeding amount can be adjusted when abnormal, so that the growth condition of beef cattle can quickly return to normal, and the growth condition of beef cattle can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and in particular to a beef cattle breeding management method and system based on machine learning. Background Art

[0002] Beef is the second largest meat food in my country, second only to pork. Compared with other livestock meats, beef has a relatively high protein content, but relatively low fat and cholesterol content. Beef is also rich in minerals and contains 8 essential amino acids for the human body in a balanced ratio. Almost all of it can be absorbed and utilized by the human body after ingestion. Its fat content is lower than that of pork and mutton. With the improvement of people's requirements for food safety and quality and the development of science and technology, traditional beef cattle breeding and management methods can no longer meet the needs of modern, efficient and high-quality production.

[0003] CN118071034B discloses a method and system for feeding and managing beef cattle, which belongs to the field of intelligent management technology. First, the feces image of beef cattle is collected and preprocessed to improve the image quality for subsequent analysis; then the preprocessed feces image is input into the feces classification network for feature extraction and classification, and the network includes an expansion convolution module, a dual-branch feature fusion module and a fully connected layer, which can deeply mine the key information in the feces image and accurately identify the feces type; at the same time, the feed trough data is collected, the features are obtained through the sensor, and the machine learning algorithm is used to analyze the feed remaining state and the feeding behavior of beef cattle; then these features are fused to construct a fusion feature set, which is input into the LSTM model for training to predict the feed trough state; finally, the feces classification results are correlated with the feed trough state evaluation results to form a comprehensive feeding management decision basis, which provides scientific and accurate decision support for beef cattle feeding and management.

[0004] CN117910709B discloses an intelligent cattle farm feeding management system, including a cattle house information collection module, a feed information collection module, an image information collection module, a protection information collection module, a data processing module and an information sending module; the cattle house information collection module is used to collect cattle house related information, the feed information collection module is used to collect feed related information; the image information collection module is used to collect image information in the cattle house, the protection information collection module is used to collect protection related information in the cattle house, and the protection related information is the protection insect cultivation information in the cattle house; the data processing module is used to process the cattle house related information, feed related information, image information in the cattle house and protection related information to generate the first management information, the second management information and the third management information. This solution can carry out intelligent cattle farm feeding management more comprehensively.

[0005] Therefore, in the related art, although it is possible to observe the feces and cowshed conditions of beef cattle based on image collection, it is impossible to directly observe the growth of beef cattle, nor is it possible to reasonably adjust the feed ratio and feeding amount.

[0006] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0007] The present invention provides a beef cattle feeding and management method and system based on machine learning, which can solve the technical problems that the related technology cannot directly observe the growth of beef cattle and cannot reasonably adjust the feed ratio and feeding amount.

[0008] According to a first aspect of the present invention, there is provided a beef cattle breeding and management method based on machine learning, comprising:

[0009] Determine the growth stage of the same batch of beef cattle in the current monitoring period;

[0010] At the beginning of the current monitoring cycle, first images of the plurality of beef cattle at multiple angles are taken, and a first weight of each beef cattle is weighed; at the end of the current monitoring cycle, second images of the plurality of beef cattle at multiple angles are taken, and a second weight of each beef cattle is weighed;

[0011] Determine the feed ratio and daily feed amount within the current monitoring period;

[0012] Processing the first image and the second image using the trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle;

[0013] Determining a growth abnormality score for each beef cattle according to the growth morphology abnormality score, the first body weight, and the second body weight;

[0014] Determining whether the feed ratio and the feed feeding amount need to be adjusted according to the abnormal growth score;

[0015] If adjustment is needed, the adjusted feed ratio and the adjusted feed amount are obtained according to the abnormal growth score, the feed ratio in the current monitoring period, the daily feed feeding amount, the number of beef cattle in the same batch, the growth stage, the end time of the current monitoring period and the trained feed optimization model;

[0016] In the next monitoring period, the adjusted feed ratio and adjusted feed feeding amount are used for feeding.

[0017] According to a second aspect of the present invention, there is provided a beef cattle feeding management system based on machine learning, comprising:

[0018] Growth stage module, used to determine the growth stage of the same batch of beef cattle in the current monitoring period;

[0019] A shooting module, used for shooting first images of multiple beef cattle at multiple angles and weighing a first weight of each beef cattle at the beginning of a current monitoring cycle, and shooting second images of multiple beef cattle at multiple angles and weighing a second weight of each beef cattle at the end of the current monitoring cycle;

[0020] A determination module is used to determine the feed ratio and daily feed amount within the current monitoring period;

[0021] A growth morphology abnormality scoring module, used to process the first image and the second image by using a trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle;

[0022] A growth abnormality scoring module, used to determine the growth abnormality score of each beef cattle according to the growth morphology abnormality score, the first body weight and the second body weight;

[0023] A judgment module, used to determine whether the feed ratio and the feed feeding amount need to be adjusted according to the abnormal growth score;

[0024] An adjustment module, for obtaining an adjusted feed ratio and an adjusted feed amount according to the abnormal growth score, the feed ratio in the current monitoring period, the daily feed amount, the number of beef cattle in the same batch, the growth stage, the end time of the current monitoring period and the trained feed optimization model if adjustment is needed;

[0025] The feeding module is used for feeding with the adjusted feed ratio and the adjusted feed feeding amount in the next monitoring cycle.

[0026] Technical effect: According to the present invention, it is possible to take images of beef cattle and weigh the weight of beef cattle at the beginning and end of the monitoring period, so as to directly observe the growth status of beef cattle, and determine whether the growth status of beef cattle is abnormal through the abnormal growth morphology recognition model. If the growth status is abnormal, the feed ratio and feed feeding amount can be adjusted to make the growth status of beef cattle quickly return to normal, improve the accuracy of monitoring, and improve the scientific nature of feeding management, reduce the probability of abnormal growth, improve the growth status of beef cattle, and increase the yield of beef cattle. When determining the reference score, the reference score in two cases can be represented by a conditional function. When the degree of deviation between the morphology of beef cattle and the standard morphology increases, the product of the increase multiple and the degree of deviation is used to represent the severity of the abnormal morphology of beef cattle, which can indicate that the morphology of beef cattle is accelerating abnormalization and the severity of the abnormal morphology of beef cattle is high. On the contrary, only the degree of deviation can be used to represent the severity of the abnormal morphology of beef cattle, indicating that the severity of the abnormal morphology of beef cattle is low, so that the severity of the abnormal morphology of beef cattle can be objectively and accurately described. When determining the first loss function, the angle information of the line between adjacent key points can be used to represent the posture of the beef cattle, thereby removing the key point position difference caused by the body shape difference, improving the accuracy of the posture representation, and the posture deviation between the corrected posture and the reference posture, as well as the deviation between the reference score and the sample growth morphology abnormality score can be used to form the first loss function, thereby improving the posture correction ability of the growth morphology abnormality recognition model and the calculation accuracy of the growth morphology abnormality score, and improving the performance of the growth morphology abnormality recognition model. When determining the growth abnormality score, the normal score of the normal morphology of the beef cattle can be determined, and the normal score of the weight growth of the beef cattle can also be determined, thereby comprehensively describing the normality of the body shape change of the beef cattle from two aspects, and then solving the abnormality of the body shape change of the beef cattle, obtaining the growth abnormality score, and accurately and objectively describing the abnormality of the body shape change of the beef cattle. When determining the second loss function, the feeding error of the feed can be determined based on the difference between the historical ratio vector and the sample ratio vector and the difference between the historical average feed feeding amount and the sample average feed feeding amount, and the urgency and importance of the adjustment, as well as the reference value and importance of the historical feed ratio and the historical average feed feeding amount in the target historical monitoring period, to determine the weight of the feeding error, improve the objectivity and accuracy of the weight, and conduct targeted training on the feed optimization model to improve the accuracy of the feed optimization model.

[0027] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other embodiments can be obtained based on these drawings without creative work.

[0029] Figure 1 A schematic diagram of a process flow of a beef cattle feeding and management method based on machine learning according to an embodiment of the present invention is exemplarily shown;

[0030] Figure 2 The beef cattle feeding and management system based on machine learning according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0033] Figure 1 A flow chart of a beef cattle feeding and management method based on machine learning according to an embodiment of the present invention is exemplarily shown, and the method comprises:

[0034] Step S101, determining the growth stage of the same batch of beef cattle in the current monitoring period;

[0035] Step S102, at the beginning of the current monitoring cycle, taking first images of the plurality of beef cattle at multiple angles, and weighing a first weight of each beef cattle, and at the end of the current monitoring cycle, taking second images of the plurality of beef cattle at multiple angles, and weighing a second weight of each beef cattle;

[0036] Step S103, determining the feed ratio and daily feed amount in the current monitoring period;

[0037] Step S104, processing the first image and the second image by using the trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle;

[0038] Step S105, determining a growth abnormality score for each beef cattle according to the growth morphology abnormality score, the first body weight and the second body weight;

[0039] Step S106, determining whether the feed ratio and the feed feeding amount need to be adjusted according to the growth abnormality score;

[0040] Step S107, if adjustment is needed, obtaining the adjusted feed ratio and the adjusted feed amount according to the growth abnormality score, the feed ratio in the current monitoring period, the daily feed feeding amount, the number of beef cattle in the same batch, the growth stage, the end time of the current monitoring period and the trained feed optimization model;

[0041] Step S108, feeding the pigs using the adjusted feed ratio and the adjusted feed feeding amount in the next monitoring period.

[0042] According to the beef cattle feeding and management method based on machine learning in an embodiment of the present invention, it is possible to take images of beef cattle and weigh the beef cattle at the beginning and end of the monitoring period, so as to directly observe the growth condition of the beef cattle, and judge whether the growth condition of the beef cattle is abnormal through the growth morphology abnormality recognition model. If the growth condition is abnormal, the feed ratio and feed feeding amount can be adjusted so that the growth condition of the beef cattle can quickly return to normal, thereby improving the accuracy of monitoring, and improving the scientific nature of feeding and management, reducing the probability of abnormal growth conditions, improving the growth condition of beef cattle, and increasing the yield of beef cattle.

[0043] According to one embodiment of the present invention, the growth of beef cattle can be divided into multiple growth stages, including the calf stage, the feeder stage, the early fattening stage, the late fattening stage, etc., and each growth stage lasts for several months or even years. And the growth rate of each stage is different, and the required feed ratio and feeding amount are also different. Therefore, in step S101, the growth stage of the current monitoring cycle can be determined, so as to determine whether the growth status of the beef cattle in this growth stage is abnormal in the subsequent processing, and determine the appropriate feed ratio and feeding amount, wherein the duration of each monitoring cycle is one month or one quarter, and the present invention does not limit the duration of the monitoring cycle.

[0044] According to one embodiment of the present invention, in step S102, at the beginning of the current monitoring cycle, first images of multiple beef cattle at multiple angles are taken, and first images of each beef cattle at angles such as the front, side, back, and top can be taken, and the first weight of each beef cattle can also be weighed. Correspondingly, at the end of the current monitoring cycle, second images of the multiple angles of the multiple beef cattle are taken, and the second weight of each beef cattle is weighed.

[0045] According to one embodiment of the present invention, in step S103, the feed ratio and the daily feed amount in the current monitoring period can be obtained, and based on the ratio of the daily feed amount to the number of beef cattle, the average feed amount for each beef cattle can be determined.

[0046] According to one embodiment of the present invention, in step S104, the growth morphology abnormality recognition model is a deep learning neural network model, for example, a convolutional neural network model, which can process the first image and the second image, and determine whether the growth morphology of the beef cattle in the first image and the second image is abnormal.

[0047] According to one embodiment of the present invention, the first image and the second image are processed by a trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle, including: inputting the first image and the second image of the same beef cattle into the encoding layer of the trained growth morphology abnormality recognition model to obtain a first feature map of the first image and a second feature map of the second image respectively; identifying the posture of the beef cattle in the first feature map through the posture recognition layer of the trained growth morphology abnormality recognition model to obtain first posture information, and identifying the posture of the beef cattle in the second feature map to obtain second posture information; correcting the first posture information to obtain third posture information, and correcting the second posture information to obtain fourth posture information; inputting the third posture information and the first feature map into the posture change layer of the trained growth morphology abnormality recognition model to obtain a third feature map after posture transformation, and inputting the fourth posture information and the second feature map into the posture change layer of the trained growth morphology abnormality recognition model to obtain a fourth feature map after posture transformation; inputting the third feature map and the fourth feature map into the decoding layer of the trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for the beef cattle.

[0048] According to one embodiment of the present invention, the abnormal growth morphology recognition model may include a coding level, a posture recognition level, a posture change level and a decoding level. Among them, the coding level can be used to perform convolution, pooling, activation and other processing on the input first image and the second image to obtain the first feature map of the first image and the second feature map of the second image, so as to improve the receptive field, so as to facilitate the observation of the morphology of the beef cattle, the proportion of each part of the body and other information. The posture recognition level can be used to identify the posture of the beef cattle in the first feature map and the second feature map, and obtain the first posture information and the second posture information respectively, wherein the first posture information and the second posture information can both be the relative position relationship between multiple feature points on the trunk of the beef cattle, and the present invention does not limit the specific types of the first posture information and the second posture information. When the beef cattle shoot the first image and the second image, the posture may not be consistent, so it is not convenient to compare, and its posture may not be suitable for observing the growth morphology of the beef cattle. Therefore, the first posture information and the second posture information can be corrected to be a reference posture suitable for observing the growth morphology, and the posture of the corrected beef cattle can also be kept consistent, which is convenient for comparison. During the calibration process, since the body shape of the beef cattle may change, even if the beef cattle maintain the reference posture, the relative positional relationship of each key point at the beginning and end of the monitoring period may still be different. Therefore, this calibration process does not calibrate the positional relationship between the key points, but calibrates the angle of the line between adjacent key points. For example, the line connecting the key points on the cow's legs and the line connecting the key points on the cow's body can be corrected to 90 degrees. The present invention does not limit the specific value of the angle. Further, the posture of the beef cattle in the feature map can be adjusted based on the corrected posture information to obtain a new feature map. This processing can be performed through the posture change level, that is, the third posture information and the first feature map are input into the posture change level, so that the posture of the beef cattle in the first feature map is corrected to obtain the third feature map. In the third feature map, the posture of the beef cattle is the above-mentioned reference posture information. Similarly, the fourth posture information and the second feature map are input into the posture change level to obtain the fourth feature map after the posture transformation. The decoding level can decode the third and fourth characteristic maps after posture correction, so as to analyze the information such as the body shape and proportions of various body parts of the beef cattle, thereby eliminating the influence of posture factors and obtaining a growth morphological abnormality score of the beef cattle with higher accuracy.

[0049] According to one embodiment of the present invention, the above-mentioned growth morphology abnormality recognition model can be trained before use, and the training steps of the growth morphology abnormality recognition model include: inputting the first sample image and the second sample image of the same beef cattle at the beginning of the historical monitoring cycle into the encoding layer of the growth morphology abnormality recognition model to obtain the first sample feature map of the first sample image and the second sample feature map of the second sample image; processing the first sample feature map and the second sample feature map through the posture recognition layer of the growth morphology abnormality recognition model to obtain the first sample posture information corresponding to the first sample feature map and the second sample posture information corresponding to the second sample feature map; correcting the first sample posture information to obtain the third sample posture information, and correcting the second sample posture information to obtain the fourth sample posture information; processing the third sample posture information and the first feature map through the posture change layer to obtain the third sample feature map, and processing the fourth sample posture information and the second feature map through the posture change layer , obtain a fourth sample feature map; decode the third sample feature map and the fourth sample feature map through a key point decoding layer to obtain a first decoded image corresponding to the third sample feature map, and a second decoded image corresponding to the fourth sample feature map; perform key point detection on the first decoded image to obtain the first key point position coordinates in the first decoded image, and perform key point detection on the second decoded image to obtain the second key point position coordinates in the second decoded image; determine a reference score based on the annotation information of the first sample image and the second sample image; input the third sample feature map and the fourth sample feature map into the decoding layer of the growth morphology abnormality recognition model to obtain the sample growth morphology abnormality score; determine a first loss function of the growth morphology abnormality recognition model based on the first key point position coordinates, the second key point position coordinates, the reference score and the sample growth morphology abnormality score; train the growth morphology abnormality recognition model based on the first loss function to obtain a trained growth morphology abnormality recognition model.

[0050] According to one embodiment of the present invention, during the training process, the ability of the growth morphology abnormality recognition model to correct posture information and the ability to calculate the growth morphology abnormality score can be trained respectively. The first sample feature map of the first sample image and the second sample feature map of the second sample image can be obtained using the encoding level, and the first sample posture information corresponding to the first sample feature map and the second sample posture information corresponding to the second sample feature map can be obtained through the posture recognition level. After correction in the above manner, the third sample posture information and the first feature map are processed through the posture change level to obtain the third sample feature map, and the fourth sample posture information and the second feature map are processed through the posture change level to obtain the fourth sample feature map. In order to train the ability of the growth morphology abnormality recognition model to correct posture information, the key point decoding level can be used to decode the third sample feature map and the fourth sample feature map to obtain a first decoded image corresponding to the third sample feature map and a second decoded image corresponding to the fourth sample feature map. The first decoded image and the second decoded image are images after the posture of the beef cattle is corrected. In the first decoded image and the second decoded image, the beef cattle should theoretically maintain the above-mentioned reference posture. However, since the growth morphology abnormality recognition model may have errors, the key point detection model can be used to detect the position coordinates of the first key point in the first decoded image and the position coordinates of the second key point in the second decoded image, so as to determine the posture of the beef cattle in the first decoded image and the second decoded image based on the position coordinates of the first key point and the position coordinates of the second key point, and then determine the loss function through the error between the posture of the beef cattle in the first decoded image and the second decoded image and the reference posture, and then minimize the loss function during the training process, thereby reducing the error between the corrected posture of the beef cattle and the reference posture.

[0051] According to one embodiment of the present invention, the ability of the growth morphology abnormality recognition model to calculate the growth morphology abnormality score can also be trained. The reference score of the beef cattle in the first sample image and the second sample image can be first calculated by annotating information as the correct growth morphology abnormality score of the beef cattle. The reference score can be compared with the sample growth morphology abnormality score output by the decoding layer to obtain the error of the growth morphology abnormality recognition model in calculating the growth morphology abnormality score, and then the loss function is determined based on the error, and then the loss function is minimized during the training process, thereby reducing the error of the growth morphology abnormality recognition model in calculating the growth morphology abnormality score and improving the accuracy of the growth morphology abnormality recognition model.

[0052] According to an embodiment of the present invention, determining a reference score based on the annotation information of the first sample image and the second sample image includes: determining a first size ratio of various parts of the beef cattle body in the first sample image based on the annotation information of the first sample image; determining a second size ratio of various parts of the beef cattle body in the second sample image based on the annotation information of the second sample image; and determining the reference score according to formula (1): ,

[0053] (1)

[0054] in, is the first size ratio vector composed of the first size ratios of each part, is the first size ratio preset vector at the beginning of the current monitoring cycle, for The transposed vector of is the second size ratio vector composed of the second size ratios of each part, The second size ratio preset vector at the end of the current monitoring period, for The transposed vector of , if is the conditional function.

[0055] According to one embodiment of the present invention, in formula (1), the conditional function Indicates that the conditions are met In the case of , otherwise, the conditional function value is .

[0056] Among them, the first dimension ratio of each part may include the ratio of the length of the cow's legs to the body, the ratio of the body width to the body length, the ratio of the length of the cow's head to the body, etc. The present invention does not limit the first dimension ratio. The first dimension ratio vector is a vector obtained by combining the above-mentioned multiple first dimension ratios. The various components in the first dimension ratio preset vector are the standard preset values ​​of the above-mentioned various ratios at the beginning of the current monitoring cycle. The definition of the second dimension ratio is similar to that of the first dimension ratio, which will not be repeated here. The various components in the second dimension ratio preset vector are the standard preset values ​​of the various ratios at the end of the current monitoring cycle. If the above conditions are met, it means that at the beginning of the current monitoring cycle, the morphology of the beef cattle is more similar to the standard morphology. In other words, after a monitoring cycle of feeding, the morphology of the beef cattle deviates more from the standard morphology, then As a reference score, The degree of deviation between the morphology of beef cattle and the standard morphology at the end of the current monitoring period. The difference between the degree of deviation between the morphology of beef cattle and the standard morphology at the end of the current monitoring period and the degree of deviation between the morphology of beef cattle and the standard morphology at the beginning of the current monitoring period. is the relative difference between the two, that is, the relative change in the degree of deviation between the morphology of beef cattle and the standard morphology during the current monitoring period. It indicates the increase in the multiple of the deviation between the morphology of beef cattle and the standard morphology during the current monitoring period. This can indicate the severity of the abnormal morphology of the beef cattle, that is, it indicates that the morphology of the beef cattle is accelerating abnormalization, and the severity of the abnormal morphology of the beef cattle is high. If the above conditions are not met, the degree of deviation between the morphology of the beef cattle and the standard morphology at the end of the current monitoring period can be used as a reference score, indicating that although there is a certain deviation between the morphology of the beef cattle and the standard morphology, the degree of deviation has not increased, and the severity of the abnormal morphology of the beef cattle is low.

[0057] In this way, the reference scores in the two cases can be represented by the conditional function. When the degree of deviation between the morphology of the beef cattle and the standard morphology increases, the product of the increase multiple and the degree of deviation is used to represent the severity of the morphological abnormality of the beef cattle, which can indicate that the morphology of the beef cattle is accelerating abnormalization and the severity of the morphological abnormality of the beef cattle is high. Conversely, only the degree of deviation can be used to represent the severity of the morphological abnormality of the beef cattle, indicating that the severity of the morphological abnormality of the beef cattle is low, thereby being able to objectively and accurately describe the severity of the morphological abnormality of the beef cattle.

[0058] According to an embodiment of the present invention, the first loss function of the growth morphology abnormality recognition model is determined according to the first key point position coordinates, the second key point position coordinates, the reference score and the sample growth morphology abnormality score, including: determining the first angle information between the lines connecting adjacent first key points according to the first key point position coordinates; determining the second angle information between the lines connecting adjacent second key points according to the second key point position coordinates; determining the first loss function of the growth morphology abnormality recognition model according to formula (2): , (2)

[0059] in, is a first angle vector composed of multiple first angle information, for The transposed vector of is the reference angle vector composed of the angle information between the lines connecting adjacent key points on the beef cattle trunk in the reference posture, is a second angle vector composed of multiple second angle information, for The transposed vector of For reference rating, Score the sample for abnormal growth morphology, , and The preset weights.

[0060] According to one embodiment of the present invention, the posture of the beef cattle can be described by the angle information of the lines between adjacent key points. The components of the first angle vector are the first angle information of the multiple lines. Similarly, the components of the second angle vector are the second angle information of the multiple lines. The components of the reference angle vector are the reference angle vectors of the multiple lines under the reference posture. Therefore, in formula (2), may represent a posture deviation between the posture of the beef cattle in the first decoded image and the reference posture, It can represent the posture deviation between the posture of the beef cattle in the second decoded image and the reference posture, and these two items can be used as items in the first loss function, so as to reduce the above two posture deviations and improve the ability of the growth morphology abnormality recognition model to correct the posture.

[0061] According to one embodiment of the present invention, in formula (2), It represents the deviation between the reference score and the sample growth morphology anomaly score calculated by the growth morphology anomaly recognition model. The deviation can be used as a term in the first loss function to reduce the deviation and improve the accuracy of the sample growth morphology anomaly score calculated by the growth morphology anomaly recognition model.

[0062] According to one embodiment of the present invention, the above three items can be weighted and summed to obtain a first loss function, so that during the training process, the first loss function is back-propagated to adjust the parameters of the growth morphology abnormality recognition model to reduce the loss function, thereby improving the ability of the growth morphology abnormality recognition model to correct the posture and the accuracy of the sample growth morphology abnormality score. The above training steps can be performed multiple times to train the growth morphology abnormality recognition model multiple times, so that the trained growth morphology abnormality recognition model can be obtained for calculating the growth morphology abnormality score.

[0063] In this way, the angle information of the line between adjacent key points can be used to represent the posture of beef cattle, thereby removing the key point position differences caused by body shape differences and improving the accuracy of posture representation. The posture deviation between the corrected posture and the reference posture, as well as the deviation between the reference score and the sample growth morphology abnormality score can be used to form the first loss function, thereby improving the posture correction ability of the growth morphology abnormality recognition model and the calculation accuracy of the growth morphology abnormality score, thereby improving the performance of the growth morphology abnormality recognition model.

[0064] According to one embodiment of the present invention, in step S105, the growth abnormality score of each beef cattle is determined according to the growth morphology abnormality score, the first body weight and the second body weight, including: determining the growth abnormality score of the i-th beef cattle according to formula (3): , (3)

[0065] in, is the second weight of the i-th beef cattle, is the first weight of the i-th beef cattle, To preset the weight gain ratio, Score the abnormal growth morphology of the i-th beef cattle.

[0066] According to one embodiment of the present invention, in formula (3), It can indicate the closeness of growth morphology to the standard morphology and can be used as a normal growth morphology score for beef cattle. is the relative change in the weight of beef cattle, is the relative difference between the preset weight gain ratio and the relative change in body weight, which can indicate an abnormal weight gain. The two are multiplied to obtain the normality of the body shape change of beef cattle in terms of morphology and weight growth, that is, the normal degree of growth. Therefore, 1 minus the product can indicate the abnormal degree of growth of the head cattle, which can be used as the abnormal growth score of beef cattle.

[0067] In this way, the normal score of the normal morphology of beef cattle can be determined, and the normal score of the weight gain of beef cattle can also be determined, so as to comprehensively describe the normality of the body shape changes of beef cattle from two aspects, and then solve the abnormality of the body shape changes of beef cattle, and obtain the growth abnormality score, which can accurately and objectively describe the abnormal situation of the body shape changes of beef cattle.

[0068] According to one embodiment of the present invention, in step S106, it can be determined whether the feed ratio and feed feeding amount need to be adjusted based on the growth abnormality score. For example, the average value of the growth abnormality scores of multiple beef cattle can be calculated. If the average value of the growth abnormality scores is lower than a preset score threshold, it is determined that the feed ratio and feed feeding amount need to be adjusted.

[0069] According to one embodiment of the present invention, in step S107, if adjustment is needed, the ratio of the daily feed amount to the number of beef cattle in the same batch can be solved to obtain the average feed amount of each beef cattle, and the average value of the growth abnormality score, the average feed amount, the feed ratio, the growth stage and the end time of the current monitoring cycle are input into the trained feed optimization model to obtain the adjusted feed ratio and the adjusted feed amount.

[0070] According to one embodiment of the present invention, the feed optimization model is a deep learning neural network model, for example, a BP neural network model, and the present invention does not limit the specific type of the feed optimization model. The feed optimization model can be trained before use, and the training steps of the feed optimization model include: obtaining the average value of the growth abnormality scores of multiple beef cattle in multiple historical monitoring cycles, and screening out the target historical monitoring cycles that are in the same growth stage as the current monitoring cycle and the average value of the growth abnormality scores of two adjacent historical monitoring cycles has a decrease greater than or equal to a preset range;

[0071] According to the daily historical feed feeding amount and the number of beef cattle in the target historical monitoring period, the historical average feed feeding amount of each beef cattle in the target historical monitoring period is determined; the average growth abnormality scores of multiple beef cattle in the j-th target historical monitoring period, the historical feed ratio in the j-th target historical monitoring period, the historical average feed feeding amount of each beef cattle in the j-th target historical monitoring period, the growth stage of the j-th target historical monitoring period and the end time of the j-th target historical monitoring period are input into the feed optimization model to obtain the sample feed ratio of the j+1-th target historical monitoring period and the sample feed average feeding amount of the j+1-th target historical monitoring period; according to the The second loss function of the feed optimization model is determined based on the average growth abnormality scores of the j target historical monitoring periods, the average growth abnormality scores of the j+1 target historical monitoring period, the historical feed ratio of the j+1 target historical monitoring period, the historical average feed feeding amount of the j+1 target historical monitoring period, the sample feed ratio of the j+1 target historical monitoring period, the sample average feed feeding amount of the j+1 target historical monitoring period, the end time of the j target historical monitoring period, and the growth stage of the j target historical monitoring period; the feed optimization model is trained according to the second loss function to obtain the trained feed optimization model.

[0072] According to one embodiment of the present invention, a target historical monitoring cycle can be screened from the historical monitoring cycles of beef cattle raised for many years, so that the data of the target historical monitoring cycle can be used for training. Since the demand for feed and the demand ratio are different in different growth stages, it is more useful to use the historical monitoring cycle with the same growth stage. Furthermore, if the decrease in the average value of the growth abnormality score of two adjacent historical monitoring cycles is greater than or equal to the preset amplitude, it means that the feed feeding amount and feed ratio are adjusted in the subsequent historical monitoring cycle, which significantly improves the growth condition of beef cattle. Therefore, the historical monitoring cycle that meets this condition can be used as the target historical monitoring cycle. For example, if the jth historical monitoring cycle and the j+1th historical monitoring cycle meet the above conditions, the jth historical monitoring cycle and the j+1th historical monitoring cycle are both used as the target historical monitoring cycle.

[0073] According to one embodiment of the present invention, the average value of the growth abnormality scores of multiple beef cattle in the jth target historical monitoring period, the historical feed ratio in the jth target historical monitoring period, the historical average feed feeding amount of each beef cattle in the jth target historical monitoring period, the growth stage of the jth target historical monitoring period and the end time of the jth target historical monitoring period can be input into a feed optimization model, and the feed optimization model can output the sample feed ratio of the j+1th target historical monitoring period and the sample average feed feeding amount of the j+1th target historical monitoring period. Since the historical average feed feeding amount and historical feed ratio of the j+1th target historical monitoring period can significantly improve the growth condition of beef cattle, the historical average feed feeding amount and historical feed ratio of the j+1th target historical monitoring period can be used as reference values. The feed optimization model can output the sample feed ratio of the j+1th target historical monitoring period and the sample average feed feeding amount of the j+1th target historical monitoring period. There may be an error between the reference value, and the second loss function can be determined based on the error, so as to train the feed optimization model to improve the calculation accuracy of the feed ratio and the average feed feeding amount.

[0074] According to one embodiment of the present invention, according to the average value of the growth abnormality score of the j-th target historical monitoring period, the average value of the growth abnormality score of the j+1-th target historical monitoring period, the historical feed ratio of the j+1-th target historical monitoring period, the historical feed average feeding amount of the j+1-th target historical monitoring period, the sample feed ratio of the j+1-th target historical monitoring period, the sample feed average feeding amount of the j+1-th target historical monitoring period, the end time of the j-th target historical monitoring period, and the growth stage of the j-th target historical monitoring period, the second loss function of the feed optimization model is determined, including: determining the second loss function of the feed optimization model according to formula (4): ,

[0075] (4)

[0076] in, is the average value of the growth abnormality score of the jth target historical monitoring period, is the average value of the growth abnormality score of the j+1th target historical monitoring period, is the end time of the jth target historical monitoring cycle, is the starting time of the growth phase of the jth target historical monitoring period, is the end time of the growth phase of the j-th target historical monitoring period, is the historical feed ratio vector composed of the historical feed ratio of the j+1th target historical monitoring period, The sample ratio vector composed of the sample feed ratio of the j+1th target historical monitoring period, is the average historical feed feeding amount in the j+1th target historical monitoring period, is the average feeding amount of sample feed in the j+1th target historical monitoring period, and is the preset weight.

[0077] According to one embodiment of the present invention, in formula (4), represents the difference between the historical matching vector and the sample matching vector, It represents the difference between the average historical feed amount and the average sample feed amount. The above two differences can be weighted and summed to get the feed feeding error, which can be used as the term of the second loss function to reduce the feed feeding error during training and improve the accuracy of the feed optimization model.

[0078] According to one embodiment of the present invention, It is the ratio of the elapsed time of the growth stage of the target historical monitoring period at the end of the jth target historical monitoring period to the total time of the growth stage. The longer the elapsed time is, the larger the ratio is, and the closer it is to the end of the growth stage. Therefore, the urgency and importance of adjusting the average feed amount and feed ratio are higher. Therefore, this ratio can be used as the weight of the above feeding error to describe the urgency and importance of adjustment.

[0079] According to one embodiment of the present invention, It is the decrease in the average value of the growth abnormality score of the j+1th target historical monitoring period relative to the average value of the growth abnormality score of the jth target historical monitoring period. The higher the decrease, the better the effect of the historical feed ratio and the historical average feed feeding amount of the j+1th target historical monitoring period. In other words, the more accurate the historical feed ratio and the historical average feed feeding amount are, the higher the reference value and importance of these data. Therefore, the decrease can be used as the weight of the upper feeding error to describe the reference value and importance of the historical feed ratio and the historical average feed feeding amount of the j+1th target historical monitoring period.

[0080] According to one embodiment of the present invention, the above two weights are multiplied to obtain the comprehensive weight of the feeding error, and the second loss function of the feed optimization model can be obtained after weighted summation using the comprehensive weights. The second loss function can be used for back propagation to adjust the parameters of the feed optimization model, and the above training steps can be iteratively performed. After multiple trainings, the trained feed optimization model is obtained to calculate the adjusted feed ratio and the adjusted average feed amount of each beef cattle, and then the adjusted feed amount is obtained by multiplying the average feed amount of each beef cattle by the number of beef cattle.

[0081] In this way, the feeding error of feed can be determined based on the difference between the historical ratio vector and the sample ratio vector and the difference between the historical average feed feeding amount and the sample average feed feeding amount. The urgency and importance of adjustment, as well as the reference value and importance of the historical feed ratio and the historical average feed feeding amount of the target historical monitoring period, can be taken into account to determine the weight of the feeding error, improve the objectivity and accuracy of the weight, and conduct targeted training on the feed optimization model to improve its accuracy.

[0082] According to one embodiment of the present invention, in step S108, the adjusted feed ratio and the adjusted feed feeding amount are used for feeding in the next monitoring period, thereby improving the growth condition of the beef cattle. If in step S106, it is determined that the feed ratio and the feed feeding amount do not need to be adjusted, the feed ratio and the feed feeding amount of the current monitoring period are used as the feed ratio and the feed feeding amount of the next monitoring period.

[0083] According to the beef cattle feeding and management method based on machine learning of the embodiment of the present invention, it is possible to take images of beef cattle and weigh the weight of beef cattle at the beginning and end of the monitoring period, so as to directly observe the growth status of beef cattle, and judge whether the growth status of beef cattle is abnormal through the abnormal growth morphology recognition model. If the growth status is abnormal, the feed ratio and feed feeding amount can be adjusted to make the growth status of beef cattle quickly return to normal, improve the accuracy of monitoring, and improve the scientific nature of feeding and management, reduce the probability of abnormal growth, improve the growth status of beef cattle, and increase the yield of beef cattle. When determining the reference score, the reference score in two cases can be represented by a conditional function. When the degree of deviation between the morphology of beef cattle and the standard morphology increases, the product of the increase multiple and the degree of deviation is used to represent the severity of the abnormal morphology of beef cattle, which can indicate that the morphology of beef cattle is accelerating abnormalization and the severity of the abnormal morphology of beef cattle is high. On the contrary, only the degree of deviation can be used to represent the severity of the abnormal morphology of beef cattle, indicating that the severity of the abnormal morphology of beef cattle is low, so that the severity of the abnormal morphology of beef cattle can be objectively and accurately described. When determining the first loss function, the angle information of the line between adjacent key points can be used to represent the posture of the beef cattle, thereby removing the key point position difference caused by the body shape difference, improving the accuracy of the posture representation, and the posture deviation between the corrected posture and the reference posture, as well as the deviation between the reference score and the sample growth morphology abnormality score can be used to form the first loss function, thereby improving the posture correction ability of the growth morphology abnormality recognition model and the calculation accuracy of the growth morphology abnormality score, and improving the performance of the growth morphology abnormality recognition model. When determining the growth abnormality score, the normal score of the normal morphology of the beef cattle can be determined, and the normal score of the weight growth of the beef cattle can also be determined, thereby comprehensively describing the normality of the body shape change of the beef cattle from two aspects, and then solving the abnormality of the body shape change of the beef cattle, obtaining the growth abnormality score, and accurately and objectively describing the abnormality of the body shape change of the beef cattle. When determining the second loss function, the feeding error of the feed can be determined based on the difference between the historical ratio vector and the sample ratio vector and the difference between the historical average feed feeding amount and the sample average feed feeding amount, and the urgency and importance of the adjustment, as well as the reference value and importance of the historical feed ratio and the historical average feed feeding amount in the target historical monitoring period, to determine the weight of the feeding error, improve the objectivity and accuracy of the weight, and conduct targeted training on the feed optimization model to improve the accuracy of the feed optimization model.

[0084] Figure 2 The beef cattle feeding management system based on machine learning according to an embodiment of the present invention is exemplarily shown, comprising:

[0085] Growth stage module, used to determine the growth stage of the same batch of beef cattle in the current monitoring period;

[0086] A shooting module, used for shooting first images of multiple beef cattle at multiple angles and weighing a first weight of each beef cattle at the beginning of a current monitoring cycle, and shooting second images of multiple beef cattle at multiple angles and weighing a second weight of each beef cattle at the end of the current monitoring cycle;

[0087] A determination module is used to determine the feed ratio and daily feed amount within the current monitoring period;

[0088] A growth morphology abnormality scoring module, used to process the first image and the second image by using a trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle;

[0089] A growth abnormality scoring module, used to determine the growth abnormality score of each beef cattle according to the growth morphology abnormality score, the first body weight and the second body weight;

[0090] A judgment module, used to determine whether the feed ratio and the feed feeding amount need to be adjusted according to the abnormal growth score;

[0091] An adjustment module, for obtaining an adjusted feed ratio and an adjusted feed amount according to the abnormal growth score, the feed ratio in the current monitoring period, the daily feed amount, the number of beef cattle in the same batch, the growth stage, the end time of the current monitoring period and the trained feed optimization model if adjustment is needed;

[0092] The feeding module is used for feeding with the adjusted feed ratio and the adjusted feed feeding amount in the next monitoring cycle.

[0093] It should be understood by those skilled in the art that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A beef cattle breeding and management method based on machine learning, characterized in that: include: Determine the growth stage of the same batch of beef cattle in the current monitoring period; At the beginning of the current monitoring cycle, first images of the plurality of beef cattle at multiple angles are taken, and a first weight of each beef cattle is weighed; at the end of the current monitoring cycle, second images of the plurality of beef cattle at multiple angles are taken, and a second weight of each beef cattle is weighed; Determine the feed ratio and daily feed amount within the current monitoring period; Processing the first image and the second image using the trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle; Determining a growth abnormality score for each beef cattle according to the growth morphology abnormality score, the first body weight, and the second body weight; Determining whether the feed ratio and the feed feeding amount need to be adjusted according to the abnormal growth score; If adjustment is needed, the adjusted feed ratio and the adjusted feed amount are obtained according to the abnormal growth score, the feed ratio in the current monitoring period, the daily feed feeding amount, the number of beef cattle in the same batch, the growth stage, the end time of the current monitoring period and the trained feed optimization model; In the next monitoring period, the adjusted feed ratio and feed amount are used for feeding; When training the abnormal growth morphology recognition model, a first sample image and a second sample image of the same beef cattle are obtained; Determining a reference score according to the annotation information of the first sample image and the second sample image includes: Determine, according to the annotation information of the first sample image, a first size ratio of each part of the body of the beef cattle in the first sample image; Determining a second size ratio of each part of the body of the beef cattle in the second sample image according to the annotation information of the second sample image; According to the formula Determine the reference score S R , where P1 is the first size ratio vector composed of the first size ratios of each part, P 1,p The first size ratio preset vector at the beginning of the current monitoring cycle, (P1) T is the transposed vector of P1, P2 is the second size ratio vector composed of the second size ratios of each part, P 2,p The second size ratio preset vector at the end of the current monitoring period, (P2) T is the transposed vector of P2, if is a conditional function, and the reference score is used to calculate the first loss function of the abnormal growth morphology recognition model; in, The degree of deviation between the morphology of beef cattle and the standard morphology at the end of the current monitoring period. The difference between the degree of deviation between the morphology of beef cattle and the standard morphology at the end of the current monitoring period and the degree of deviation between the morphology of beef cattle and the standard morphology at the beginning of the current monitoring period. It is the relative change in the degree of deviation between the morphology of beef cattle and the standard morphology during the current monitoring period.

2. The beef cattle breeding and management method based on machine learning according to claim 1, characterized in that: The first image and the second image are processed by the trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle, including: Inputting a first image and a second image of the same beef cattle into the encoding layer of the trained growth morphology abnormality recognition model to obtain a first feature map of the first image and a second feature map of the second image respectively; Recognize the posture of the beef cattle in the first feature map through the posture recognition level of the trained growth morphology abnormality recognition model to obtain first posture information, and recognize the posture of the beef cattle in the second feature map to obtain second posture information; Correcting the first posture information to obtain third posture information, and correcting the second posture information to obtain fourth posture information; Input the third posture information and the first feature map into the posture change level of the trained growth morphology abnormality recognition model to obtain the third feature map after posture transformation, and input the fourth posture information and the second feature map into the posture change level of the trained growth morphology abnormality recognition model to obtain the fourth feature map after posture transformation; The third characteristic graph and the fourth characteristic graph are input into the decoding layer of the trained growth morphology abnormality recognition model to obtain the growth morphology abnormality score of the beef cattle.

3. The beef cattle breeding and management method based on machine learning according to claim 2, characterized in that: The training steps of the growth morphology abnormality recognition model include: Inputting a first sample image and a second sample image of the same beef cattle at the beginning of a historical monitoring period into the encoding level of the growth morphology abnormality recognition model to obtain a first sample feature map of the first sample image and a second sample feature map of the second sample image; The first sample feature map and the second sample feature map are processed through the posture recognition level of the growth morphology abnormality recognition model to obtain first sample posture information corresponding to the first sample feature map and second sample posture information corresponding to the second sample feature map; Correcting the first sample posture information to obtain third sample posture information, and correcting the second sample posture information to obtain fourth sample posture information; Processing the third sample posture information and the first feature map through the posture change level to obtain a third sample feature map, and processing the fourth sample posture information and the second feature map through the posture change level to obtain a fourth sample feature map; Decoding the third sample feature map and the fourth sample feature map through a key point decoding level to obtain a first decoded image corresponding to the third sample feature map and a second decoded image corresponding to the fourth sample feature map; Performing key point detection on the first decoded image to obtain position coordinates of first key points in the first decoded image, and performing key point detection on the second decoded image to obtain position coordinates of second key points in the second decoded image; Inputting the third sample characteristic graph and the fourth sample characteristic graph into a decoding layer of a growth morphology abnormality recognition model to obtain a sample growth morphology abnormality score; Determine a first loss function of a growth morphology abnormality recognition model according to the first key point position coordinates, the second key point position coordinates, the reference score, and the sample growth morphology abnormality score; The abnormal growth morphology recognition model is trained according to the first loss function to obtain a trained abnormal growth morphology recognition model.

4. The beef cattle breeding and management method based on machine learning according to claim 3 is characterized in that: Determining a first loss function of a growth morphology abnormality recognition model according to the first key point position coordinates, the second key point position coordinates, the reference score, and the sample growth morphology abnormality score includes: Determine first angle information between lines connecting adjacent first key points according to the position coordinates of the first key points; Determine second angle information between lines connecting adjacent second key points according to the position coordinates of the second key points; According to the formula Determine the first loss function L1 of the growth morphology abnormality recognition model, where A1 is a first angle vector composed of multiple first angle information, (A1) T is the transposed vector of A1, A R is a reference angle vector composed of angle information between lines connecting adjacent key points on the trunk of the beef cattle in the reference posture, A2 is a second angle vector composed of multiple second angle information, (A2) T is the transposed vector of A2, S R is the reference score, S S Score the abnormal growth morphology of the sample, and α1, α2, and α3 are preset weights.

5. The beef cattle breeding and management method based on machine learning according to claim 1, characterized in that: Determining the growth abnormality score of each beef cattle according to the growth morphology abnormality score, the first body weight and the second body weight, comprising: According to the formula Determine the growth abnormality score S of the i-th beef cattle AB,i , where W 2,i is the second weight of the i-th beef cattle, W 1,i is the first weight of the i-th beef cattle, K p is the preset weight gain ratio, S F,i Score the abnormal growth morphology of the i-th beef cattle.

6. The beef cattle breeding and management method based on machine learning according to claim 1, characterized in that: The training steps of the feed optimization model include: Obtain the average value of growth abnormality scores of multiple beef cattle in multiple historical monitoring cycles, and screen out a target historical monitoring cycle that is in the same growth stage as the current monitoring cycle and in which the average value of growth abnormality scores of two adjacent historical monitoring cycles decreases by a greater than or equal to a preset range; According to the daily historical feed feeding amount and the number of beef cattle in the target historical monitoring period, determine the historical average feed feeding amount of each beef cattle in the target historical monitoring period; The average value of the abnormal growth scores of multiple beef cattle in the jth target historical monitoring period, the historical feed ratio in the jth target historical monitoring period, the historical average feed feeding amount of each beef cattle in the jth target historical monitoring period, the growth stage of the jth target historical monitoring period, and the end time of the jth target historical monitoring period are input into the feed optimization model to obtain the sample feed ratio of the j+1th target historical monitoring period and the sample feed average feeding amount of the j+1th target historical monitoring period; Determine the second loss function of the feed optimization model according to the average value of the growth abnormality score of the j-th target historical monitoring period, the average value of the growth abnormality score of the j+1-th target historical monitoring period, the historical feed ratio of the j+1-th target historical monitoring period, the historical average feed feeding amount of the j+1-th target historical monitoring period, the sample feed ratio of the j+1-th target historical monitoring period, the sample feed average feeding amount of the j+1-th target historical monitoring period, the end time of the j-th target historical monitoring period, and the growth stage of the j-th target historical monitoring period; According to the second loss function, the feed optimization model is trained to obtain a trained feed optimization model.

7. The beef cattle breeding and management method based on machine learning according to claim 6, characterized in that: According to the average value of the growth abnormality score of the j-th target historical monitoring period, the average value of the growth abnormality score of the j+1-th target historical monitoring period, the historical feed ratio of the j+1-th target historical monitoring period, the historical average feed feeding amount of the j+1-th target historical monitoring period, the sample feed ratio of the j+1-th target historical monitoring period, the sample feed average feeding amount of the j+1-th target historical monitoring period, the end time of the j-th target historical monitoring period, and the growth stage of the j-th target historical monitoring period, the second loss function of the feed optimization model is determined, including: According to the formula Determine the second loss function L2 of the feed optimization model, where S AB,ave,j is the average value of the growth abnormality score of the jth target historical monitoring period, S AB,ave,j+1 is the average value of the growth abnormality score of the j+1th target historical monitoring period, T j,end is the end time of the jth target historical monitoring cycle, T start is the starting time of the growth phase of the jth target historical monitoring period, T end is the end time of the growth phase of the jth target historical monitoring period, R j+1,his is the historical feed ratio vector composed of the historical feed ratio of the j+1th target historical monitoring period, R j+1,sam The sample ratio vector composed of the sample feed ratio of the j+1th target historical monitoring period, M j+1,his is the average historical feed feeding amount in the j+1th target historical monitoring period, M j+1,sam is the average feeding amount of sample feed in the j+1th target historical monitoring period, and γ1 and γ2 are preset weights.

8. A beef cattle feeding management system based on machine learning for executing the method according to any one of claims 1 to 7, characterized in that: include: Growth stage module, used to determine the growth stage of the same batch of beef cattle in the current monitoring period; A shooting module, used for shooting first images of multiple beef cattle at multiple angles and weighing a first weight of each beef cattle at the beginning of a current monitoring cycle, and shooting second images of multiple beef cattle at multiple angles and weighing a second weight of each beef cattle at the end of the current monitoring cycle; A determination module is used to determine the feed ratio and daily feed amount within the current monitoring period; A growth morphology abnormality scoring module, used to process the first image and the second image by using a trained growth morphology abnormality recognition model to obtain a growth morphology abnormality score for each beef cattle; A growth abnormality scoring module, used to determine the growth abnormality score of each beef cattle according to the growth morphology abnormality score, the first body weight and the second body weight; A judgment module, used to determine whether the feed ratio and the feed feeding amount need to be adjusted according to the abnormal growth score; An adjustment module, for obtaining an adjusted feed ratio and an adjusted feed amount according to the abnormal growth score, the feed ratio in the current monitoring period, the daily feed amount, the number of beef cattle in the same batch, the growth stage, the end time of the current monitoring period and the trained feed optimization model if adjustment is needed; The feeding module is used for feeding with the adjusted feed ratio and the adjusted feed feeding amount in the next monitoring cycle.

Images