Text: Bread Clip Knowledge
Edit|Bread Clip Knowledge
«——【Preamble】】 ——»
In recent years, the mainland meat sheep industry has developed rapidly, and has gradually transformed and upgraded to the direction of scale and intensification, and is the pillar industry of animal husbandry economy.
At the same time, due to the characteristics of low fat and high protein and unique flavor, mutton is favored by consumers in the meat market, so that the demand for mutton in the livestock and poultry market is also growing.
However, most of the large-scale farms in mainland China are relatively backward in terms of facilities and equipment, and they mainly rely on manual management of farms for industrial management, and fail to establish an efficient and intensive modern production and management model.
Therefore, the application of smart pasture management technology in modern large-scale sheep breeding will make the breeding management of farms enter a new era, and realize the automation, informatization and intelligence of the management model.
In modern intelligent breeding, computer vision communication technology has become a research hotspot for many experts and scholars to replace human observation and measurement of animals without interfering with the normal behavior of animals.
However, a farm in Beijing uses computer vision communication technology to improve the accuracy of sheep identity, behavior status, biometric measurement and carcass segmentation recognition.
Moreover, there is no contact with individual sheep, which will not cause sheep stress and personal injury, save feeding costs, and provide certain technical support for the intelligent and precise management of meat sheep farms.
«——【Identification·Identification·】——»
Livestock and poultry identification is an important part of farm production management, which has an important impact on the control of production performance. For sheep farms, sheep identification is the traceability guarantee of livestock products such as mutton, goat milk, and wool.
It occupies an important position in the sheep growth record system, such as weight and body size changes, breeding and breeding, pregnancy status monitoring and other breeding tasks need to be identified quickly and accurately.
At present, in the intelligent monitoring and management of large-scale sheep farms on the market, most of the identification methods use embedded chips, ear incisions, electronic ear tags, etc.
These logos are not only easy to lose, but also may cause stress response and physical damage to sheep, affecting the performance of meat sheep, thus causing certain economic losses to the farm.
In order to improve the feeding management efficiency and welfare of meat sheep farms, a prototype computer vision system was established on the basis of convolutional neural network, and the results showed that the accuracy of the system for identity recognition could reach 95.8%.
Because light affects facial edge features, sheep identification will be affected to a certain extent. Zhang et al. [5] used the training of weak edge feature fusion network and backbone feature extraction network to fuse the extracted weak edge features into CNN, which significantly improved the accuracy of sheep face recognition.
The data model test shows that the recognition accuracy of the network model after fusing weak edge features reaches 92.45%, 98.15% and 94.67%, respectively.
Due to the similarity of the faces of the sheep in the farm enclosure, the long-distance recognition has a sheep face image captured by the monitoring equipment as the model test database, and a lightweight sheep face recognition model ECAS-MFC is constructed.
After testing, it is concluded that the ECAS-MFC model is lighter and has a better recognition accuracy for individual sheep recognition.
The FasterR-CNN object detection algorithm was used to detect and locate the sheep's face in the image, and the detected face images were input into 7 different classification models, and the average recognition accuracy reached 95%.
«——【·Biometric Measurement·】——»
Biometric measurement is used to define various characteristics of animals, body size and body weight are an important basis for animal scientific research and breeding, and are also important indicators to measure the production performance of meat sheep.
Manual measurement of meat sheep weight and body shape measurement is prone to errors, which will also cause stress to meat sheep and affect the growth and development of meat sheep. and the risk of zoonotic disease transmission.
Body size data can not only reflect the relationship between the body condition, carcass structure, growth and development of meat sheep and the development of various tissues and organs, but also be a key indicator to evaluate the production performance of meat sheep.
Using the image processing model of Mask-R-CNN, the images of the collected Uzhumuqin sheep were preprocessed by Mask-R-CNN data, and the obtained images were brought in by background removal, de-drying and smoothing, optimal contour extraction, and numerical operation, so as to obtain more accurate body size data.
The measurement results of sheep body size using computer-aided VIA system showed that the accuracy of VIA system measurement was more than 90%, and the error was within 3%.
Since the sheep in the pen only have various postures, which has a certain interference with the body size measurement, the charter [3] uses the support vector machine (SVM) to classify the sheep posture (normal, bow, twist the head), and then calculate the body size parameters of the sheep according to the contour feature points, and the body size data of the sheep can be quickly (8~10s) and accurately measured through the data model.
Since the wool thickness will have a great impact on the body size of sheep in the process of measuring the body size of the visual communication data model, a coefficient-weighted body scale correction algorithm based on the difference between the wool thickness at the neck and the anterior edge of the shoulder blade was proposed.
Body weight is an important performance indicator of sheep production, which is crucial to farm management decisions and the economic benefits of sheep management.
Three weight prediction models, namely Support Vector Machine Regression (SVR), Partial Least Squares Prediction (PLS) and BP Neural Network Model, were used to measure the weight of sheep, and it was found that the weight prediction model established by SVR was more accurate.
At present, for sheep weight data, the body size data of sheep is obtained by computer recognition technology, and then the regression model of weight value and body size parameters is established, and then the weight of sheep is estimated, which is the calculation method that is more commonly used at present.
The body size of the sheep was measured by visual image processing technology, and the weight of the sheep was calculated by measuring the body size data.
The body length, body height, chest circumference, chest width and other body size parameters of sheep were calculated by image data processing model, and a multiple linear regression model was established to obtain the weight estimation data of sheep with an average error of 4.1%.
Through the regression model data analysis method, the effects of body size-related parameters on the body weight of seven sheep were studied and analyzed, and the important parameters that determine the weight of sheep were proved.
«——【·Behavior Recognition·】——»
In the process of daily feeding and management, most of the time the abnormal changes in the physical state of the sheep can not be detected in time, especially in the event of diseases, the sheep's body and daily activities will be abnormal.
For example, a farm in Hebei Province uses this new technology to detect long-term lying down, decreased feed intake, moaning, etc., and these behavioral states and characteristics can help determine whether there are abnormalities in the animal's body.
In order to realize the real-time monitoring of the body and behavior of meat sheep, the captured behavior images were tested through the established CNN model, and the results showed that the model could accurately identify the four behaviors of dairy goats: standing, lying down, curling up and feeding, and the recognition rate was ideal.
The results show that the accuracy of the model is 90.30% for feeding behavior, 94.16% for standing behavior, and 91.90% for sitting and lying behavior.
For sheep in poor physical condition, an automatic tracking algorithm was used to extract the skin temperature (°C) of the head calibration feature of the sheep from infrared thermal video (IRTV). Perform two parametric engineering procedures from RGB video to evaluate heart rate (HR), beats per minute (BPM),
Respiratory rate (RR) and breaths per minute (BrPM) to assist in the automatic assessment of heat stress, and the collected sheep face images were annotated and processed for feature extraction, and the physical pain level of sheep was automatically assessed through the extracted facial expressions.
The histogram and support vector machine were used to classify the pain degree of sheep, which provided some help for the identification of sheep's physical state and disease identification and treatment.
In order to identify the prenatal behaviors of ewes (standing, walking, lying and plowing) in a timely and accurate manner, the clustering algorithm was used to identify the lying behaviors, and the recognition rate reached 99.2%.
On this basis, the BP neural network was used to identify the standing, walking and plowing behaviors, and the recognition rates were 85.7%, 88.9% and 80.0%, respectively, which improved the accuracy of the algorithm for prenatal behavior recognition of ewes.
«——【·Carcass Cutting Identification·】——»
Rapid and accurate classification and grading of sheep carcasses is the key to determining the value and quality characteristics of the carcass, and all parts of the meat sheep carcass are interconnected.
It has similar features and is difficult to identify and detect, but with the development of image semantic segmentation technology based on deep learning, this technology can be explored for real-time recognition of sheep carcass.
The high-resolution image adversarial network was used to generate a high-resolution image adversarial network, and the most suitable network model BigGAN for sheep skeleton image generation was selected, and the sheep skeleton image dataset was expanded.
According to the basic theory and structural characteristics of the image semantic segmentation network Deeplabv3+, the image detection speed of the key part recognition model of sheep carcass was optimized, and the carcass segmentation path was designed, and the automatic control system for sheep carcass segmentation was designed to achieve the expected sheep carcass segmentation task.
In view of the difficulty of mechanized segmentation of sheep skeleton, the control program based on ladder diagram was developed by using TIAPORTALV15.1, and the human-computer interaction interface was designed. According to the needs of the cutting accuracy of meat sheep carcass, a skeleton segmentation vision system was designed, and an image processing program was developed based on Open CV.
The average segmentation loss rate of the sheep skeleton segmentation line is 1.26%, the theoretical segmentation efficiency is 100/h, and the average segmentation error is 3.65mm. This carcass segmentation recognition has also been verified by data in other animals.
A non-destructive automatic classification system for cattle carcasses was established by combining sensor network system, image standardization, color correction, Canny edge detection and machine learning.
Shaanxi Science Research Institute. An intelligent pig body segmentation system was successfully developed, which carried out laser scanning on the pig carcass to be divided to obtain the image of the carcass.
After treatment, the robot is controlled to cut the carcass, which can make specific cuts to different parts of the pig carcass, and can also process different forms of pig carcass.
«——【·Conclusion·】——»
The application of computer vision system (CVS) in digital and intelligent precision sheep breeding and high-throughput phenotyping is a promising research field.
Continuously measuring traits through CVS can reduce administrative costs and optimize decision-making in the livestock industry.
But there are still challenges that need to be studied in depth and ways to connect the growing number of devices used for different applications. This enables more complex prediction algorithms based on multiple inputs and multiple outputs (joint prediction of multiple features) with a large amount of data to be tested.
Applications need to be developed to transmit the information generated by CVS to connected systems to provide more valuable information to farms.