Real-time monitoring of dairy cow activity and feeding behavior based on computer vision

Benicio, Luana Maria

Permalink

https://hdl.handle.net/2142/122125 Copy

Description

Title

Real-time monitoring of dairy cow activity and feeding behavior based on computer vision

Author(s)

Benicio, Luana Maria

Issue Date

2023-12-05

Director of Research (if dissertation) or Advisor (if thesis)

Condotta, Isabella

Committee Member(s)

• Green-Miller, Angela
• Cardoso, Felipe

Department of Study

Animal Sciences

Discipline

Animal Sciences

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

Computer Vision, Dairy Cow, Intake

Abstract

A recent challenge for dairy farmers has been the increased demand for dairy products produced in accordance with animal welfare standards. They must consider how to sustainably and profitably increase production while also promoting the welfare and health of their herds. However, as the dairy production chain intensifies, the supervision of each cow becomes increasingly complex, requiring a lot of human labor and time resources for continuous monitoring. However, automated systems, especially those that employ computer vision, can be implemented to identify and execute routine activities, reducing labor-intensive efforts. Computer vision technology offers a continuous monitoring solution covering health, behavior, environment, and nutrition. This research aims to develop an automated and continuous system using computer vision techniques to monitor individual feeding activity in dairy cows, including visits to feeders and drinkers. In addition, the aim is to generate a model for estimating dry matter intake (DMI) by analyzing depth images, allowing real-time access to individual DMI data. The experiment was conducted at the Lincoln Avenue Dairy Farm at the University of Illinois Urbana-Champaign, Urbana – IL. For the first paper of this study, four Night Owl® surveillance cameras were installed in four pens containing 25 multiparous Holstein cows. Videos were collected over 48 hours. Random videos of each pen were selected, and 1,000 images were extracted from each pen, totaling 4,000. Bounding boxes were created around all the animals in the barns, with the postures "Standing and Lying” through the Label Studio software. The YOLO object detection model was developed in Python to detect the animal's posture. Finally, a Python script was created using PyCharm to monitor the activity level of cows inside pens over a 48-hour period. In the second paper, nine multiparous Holstein Dairy cows were used. Depth images from the feeding area were collected for 24 hours over seven days, while daily dry matter intake (DMI) was collected by weighing the leftovers in each feeder and calculating the DMI on a DM (dry matter) basis. Kinect® version 2 was set up above feeders. An algorithm was written in MATLAB software (2022b) to acquire feed dimensions automatically. A Neural Network model was developed to predict the DMI based on feeder input features extracted from depth images (volume, width, length, area) and DMI manually collected as the output. The results of the dairy cow posture detection model, which uses the YOLO v8 object detection model, showed an average precision of 99.93%. However, in the final algorithm for detecting the activity level of dairy cows, it is still necessary to adjust the code's ability to distinguish the activities associated with visits to the feeders and the presence of the animal in other areas of the pen and to adopt the function of individual and continuous monitoring. The model showed promising performance for dry matter intake prediction, achieving an R2 of 0.8416 in estimating dry matter intake (DMI), with an average error of 5.47%.

Graduation Semester

2023-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/122125

Copyright and License Information

Copyright 2023 Luana Benicio

Owning Collections