Source: VIRGINIA POLYTECHNIC INSTITUTE submitted to NRP
DSFAS PARTNERSHIP: ACOUSTIC DATA AS A NOVEL TRAIT TO MANAGE WELFARE AND ENVIRONMENTAL IMPACT IN PRECISION COW FARMING
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
Reporting Frequency
Annual
Accession No.
1030530
Grant No.
2023-67021-40009
Cumulative Award Amt.
$649,741.00
Proposal No.
2022-11638
Multistate No.
(N/A)
Project Start Date
Sep 1, 2023
Project End Date
Aug 31, 2028
Grant Year
2023
Program Code
[A1541]- Food and Agriculture Cyberinformatics and Tools
Recipient Organization
VIRGINIA POLYTECHNIC INSTITUTE
(N/A)
BLACKSBURG,VA 24061
Performing Department
(N/A)
Non Technical Summary
The aim of this project is to leverage the use of acoustic data to address two significant challenges to the dairy industry: animal welfare and climate change. The long-term objective of this proposal is to develop an acoustic, data-driven tool to leverage animal welfare and accelerate the mitigation of methane emissions in precision livestock farming (PLF). The tool is an automatic and comprehensive pipeline that manages, analyzes, and evaluates production quality using acoustic data in the cow product industry. To support the long-term goal, the specific objectives of this proposal include: (1) to test acoustic data to assess calf stress and welfare post-weaning objectively, (2) to test acoustic data to determine the potential genetic variation and nutritional effects on eructation patterns to leverage future methane emission mitigation strategies, and (3) to improve the efficiency in managing acoustic data for precision livestock farming, which contains millions of data points for a single cow in one day, by developing both a Python library tool and a web-based application. This project specifically addresses two of the six Program Area Priorities of Agriculture and Food Research Initiative (AFRI), which are (B) Animal health and production and animal products and (D) Bioenergy, natural resources, and environment.
Animal Health Component
40%
Research Effort Categories
Basic
30%
Applied
40%
Developmental
30%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
3153410106033%
3083410106034%
9036099303033%
Goals / Objectives
The long-term goal of this proposal is to develop an acoustic, data-driven tool to leverage animal welfare and accelerate the mitigation of methane emissions in precision livestock farming (PLF). The tool is an automatic and comprehensive pipeline that manages, analyzes, and evaluates production quality using acoustic data in the cow product industry. To support the long-term goal, the specific objectives of this proposal include the following: (1) To test acoustic data to assess calf stress and welfare post-weaning objectively, (2) To test acoustic data to determine the potential genetic variation and nutritional effects on eructation patterns to leverage future methane emission mitigation strategies, and (3) To improve the efficiency in managing acoustic data for precision livestock farming,which contains millions of data points for a single cow in one day by developing a Python library tool and a web-based application.
Project Methods
The project consists of two components: animal management and data analytics. Co-PI Ferreira will conduct experiments using a rigorous experimental design to avoid biased observations and confounding factors in the trials. PI Chen will help collect data, including acoustic recordings and other indicators of animal welfare and methane emissions. After the trials are completed, PI Chen will continue to build software infrastructure and implement deep learning algorithms to extract meaningful information from the collected data.EffortsEfforts to deliver science-based knowledge to our target audience include extension components ofworkshops, extension, and outreach programs. We will develop innovative teaching methodologiesto effectively communicate our findings.EvaluationThe success of our project will be evaluated using several performance metrics. Specifically, we will measure the precision and recall of our model in categorizing animal emotion status and identifying eructation sound events. Our goal is to achieve a precision and recall of at least 0.8 for both metrics. This will demonstrate the effectiveness of our approach in accurately interpreting animal emotions and monitoring methane emissions.

Progress 09/01/23 to 08/31/24

Outputs
Target Audience:Throughout this reporting period, our group carried out two animal studies to collect audio data from beef cattle and dairy cows, respectively. The audio data will be disseminated in the form of a publication and an open-access dataset, providing valuable resources for scientists who are studying animal vocalizations and sounds to monitor animal behaviors. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Two students were involved in both studies. They were trained in areas of animal handling and audio data processing. The students were also trained in the use of the audio processing pipeline, which includes the use of the `librosa` library for audio processing and the `transformers` library for model training. Two students were involved in both studies. They were trained in areas of animal handling and audio data processing. The students were also trained in the use of the audio processing pipeline, which includes the use of the `librosa` library for audio processing and the `transformers` library for model training. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?We plan to publish the audio modeling pipeline for cow eructation and disseminate the dataset to the public. We also aim to finish the remaining eight animal cohorts in the second study and finish drafting the manuscript for the second study.

Impacts
What was accomplished under these goals? During the reporting period, we conducted two animal studies. The first study involved using microphones to record the eructation, grazing, and rumination sounds of beef cattle. We developed an analytical pipeline to process the audio data, which includes the following steps: 1) clipping the audio data into 5-second segments, 2) extracting Mel-frequency cepstral coefficients (MFCC) from the audio as predictive features, and 3) training a transformer-based model to predict the occurrence of the events of interest. The second study focused on using microphones to monitor the vocalizations of cows and their calves. These vocalizations were considered a proxy for weaning-induced stress, which will be correlated with cortisol levels in blood samples. To date, the second study has been completed for two out of the ten animal cohorts.

Publications