Performing Department
(N/A)
Non Technical Summary
Non-technical summarySows give birth to 15-20 piglets per litter, which is more piglets than she has functional teats. This is critical to both animal welfare and farm economic viability because when piglets outnumber teats, up to 1 in 4 piglets are at risk of dying before weaning. Furthermore, piglets will settle permanently on a specific teat after 2-3 days of age, and any teat not selected at that time will become non-functional, further increasing piglet mortality.Thus, our goal is to develop a method to automatically track piglets' nursing patterns, teat choices, and sow behavior.To achieve this goal, we will use artificial intelligence to develop an automated computer system to follow piglet movement and nursing activity. We will associate our findings with predicting at-risk piglets and develop innovative educational tools for farmers and animal caretakers.The results of this work will be a set of tools available to train farmers, animal caretakers, and students to identify specific piglets that are at risk of pre-weaning death or poor growth.The primary impact of our results will be decreased piglet preweaning mortality and improved animal welfare. Furthermore, animal caretakers will be better trained in their work, which will enhance their job satisfaction and reduce labor turnover. This all leads to improved economic outcomes for farmers by increasing their productivity and reducing farm labor costs.
Animal Health Component
0%
Research Effort Categories
Basic
50%
Applied
50%
Developmental
(N/A)
Goals / Objectives
In this project, our goal isto develop a system to improve the management of piglets by leveraging imaging systems and modern artificial intelligence (AI) technologies to automate the labor-intensive task of identifying at-risk piglets early in their life. This project aims to develop precision livestock automated systems tomonitor piglet colostral consumption, growth, and activity during the critical stages of their early post-natal life. The second is to monitor sow activity, colostrum, and milk yield and quantify the productivity of individual teats. Our third goal is toaddress piglet care protocols and develop educational materials that engage farm employee expertise. Finally, to identify barriers to caretakers' utilization of precision livestock automated systems and demonstrate animal agriculture to the public through images and reels of piglet behavior.
Project Methods
Methods:Task 1(a) study will include 20 sows selected for teat morphology, parity, and body condition. Approximately 240 piglets will be monitored and identified at birth (T0) by birth order. Piglets will be dried off, and dimensions (length, height, width, and weight) will be measured on T0, days 3, 7, 10, and 21. Piglets will be identified using both a two-colored tag system for camera identification and wax crayons on the rump of the piglets for observer cross-referencing for birth order and confirmation of first colostral intake. Once each piglet has been recorded for suckling colostrum, we will collect colostrum from glands 3, 4, and 5 for IgG. Piglets will be venous sampled at the T0 and day one for serum IgG. Sow glandular volume (depth and width of the udder) and piglet weights will be taken to assess piglet milk volume produced and consumption. Glandular volume is measuring individual mammary glands on days 1 and 16. Dimensions of one anterior, middle, and posterior gland will be recorded using a tape measure to record the distance between teats and the distance from the base of the teat to the peritoneum. Milk production will be measured on days 7 and 19-21 in three piglets per litter, with another piglet serving as an internal control using the deuterium oxide dilution (D2O) method. Piglets are sampled at 1 hour and 48 hours post-induction of D2O.Task 1(b) Validate factors on a larger dataset and varied litter sizes. In both the US and RoI, we will use an experimental design of excess piglet demand relative to teat number. A total of 120 sows* across 4 production batches (blocks) will be used, with equal groups for primiparous (PI, n=60) and multiparous (P2-4, n=60) sows. Sows will be selected for teat condition per Task 1a. Within a parity block, sows will be allocated to one of two treatments: Adequate (A) litters, in which the number of piglets will match the number of functional teats, or inadequate (IA), in which the number of piglets will outnumber functional teats by a target average of 2 piglets. We will acquire several measurements to assess the capacity of the automated system to identify piglets with suboptimal colostrum/milk intake. Piglet dimension and IgG sampling are replicated from 1(a).Task 2 (a)The first step in monitoring piglet activities is visual tracking of each piglet. Tracking will involve piglet detection and posture estimation within each frame, along with association from frame to frame and association of tracks through occlusions. We will use a two-pronged approach to maximize the likelihood of success. In the USA we will use a keypoint-based posture estimator and tracker, building on previous work. The work in Northern Ireland will use a transformer approach to improve robustness to severe occlusions of piglets common in litter. A unique two-colour tag ID will be assigned to each piglet. To manage occlusions from pen bars, three synchronized and calibrated cameras will be placed over each sow, a top-down view and two inclined views, positioned so that at least one camera can see regions occluded by the bars in the other views.Task 2(b) We will accumulate visual cues for piglet behavior and develop an at-risk predictor based on piglet time history. The piglet posture, specified as keypoint locations on the piglet relative to the sow, will be classified once per second. The time history will include behavior classification, namely lying, moving, nursing, or standing. Using this information, we will forecast piglet outcomes such as mortality or removal to identify at-risk piglets.Task 3(a), we will compare litter weight gain to early video-based tracks by using keypoint-based postures categorized into behaviors such as lying, sitting, kneeling, sternal recumbency, laying out, eating, and drinking per 24-hour period. These behaviors will be compared to piglet weights and mortality. We will develop and use a trained teat detector augmented with a regression head that predicts production through estimation of teat volume.?Task 4(a): To elevate knowledge and information dissemination, an industry-relevant Steering Committee of genetic suppliers, veterinarians, education and communication specialists, ethologists, and farm managers will be established. This committee will meet in person for assembled input and in Year 3 to test education modules.Task 4(b) will use a three-fold education approach: PigLetWatch, Sow Lactation-AIInternship and Day One Care using images and videos charting individual piglets of their survival strategies in their first 7 days in a digitized learning format. In PigLetWatch, volunteer students will have the responsibility of monitoring birth order, drying off piglets, and starting them on the heat mat. Sow Lactation-AI Internship will be the invitation of 4 students for a 3-week internship that includes online didactic content followed by 2 weeks of hands-on training at MSU-STRC in piglet Day One care. Day One Caretaking training is a learner-centered approach and will include imagery that shows caretakers how to use outcomes from visual technology to better care for all piglets born. Instructional design principles for adult learning will be used to ensure the modules are engaging and memorable, allowing for both knowledge gain and skill-building to solve problems with the litter. The 20-minute modules will be game-like in that they can be "played" simultaneous. The training program content will be delivered in three modules with content derived from conclusions from Objectives 1-3 and delivered using virtual technology education modules. The modules will be designed to cover piglet survivability risk factors in the form of case studies utilizing real-life situations that caretakers may encounter on the farm. Some case study examples will include monitoring environmental conditions, evaluating piglet and sow behavior, and identifying compromised piglets that require human intervention. Each module will contain five case studies that will provide interactive opportunities for caretakers to actively participate and make decisions based on the information provided and their experience of risk factors throughout the training. Feedback will be provided for each case study to allow learners to understand the appropriateness of their decisions, and they will have the opportunity to go back and select another, more appropriate choice. While a registered person can take Day One Caretaker, we will offer training and assessment for farm caretakers. Task 4b3. Willingness to Play: This survey study will identify cultural, emotional, and educational barriers that influence the caretaker's interest in working with sensors to improve job performance. Individuals who enroll in the training will be selected to participate in a focus group using a purposive sampling approach (n=8 focus groups; 8-10 participants/group (Morgan, 1998). Focus groups will allow participants to freely express their own opinions and generate ideas from interactions with others. Focus groups will be conducted in both Spanish and English (n=3 Spanish; n=3 English in North Carolina; n=1 Spanish; n=1 English in Michigan/Indiana). Focus groups will identify barriers that may hinder utilization of technology on-farm from a caretaker perspective. Task 4(c) PigLetArt is a collection of images using the video reels and pixel 2D images of piglets behavior derived from our research to demonstrate agriculture to the public. The material will be exhibited at Land Grant Universities such as Michigan State University, North Carolina State University, University of Nebraska, Lincoln, and with our partners at Queen's University Belfast. People who attend the exhibit can play the Virtual Education Modules to "save piglets." The number of registrants or players will be tracked through an events management system.