Adaptive high-distance RGB imaging for accurate dairy cow feed intake estimation

Area of Science:

• Agricultural Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Accurate monitoring of feed consumption is crucial for dairy farm management and optimizing animal nutrition.
• Existing methods for feed amount estimation can be labor-intensive, costly, or lack precision.

Purpose of the Study:

• To develop an intelligent, cost-effective system for estimating the total feeding amount of dairy cows using Red-Green-Blue (RGB) images.
• To provide ranch management with an efficient and accurate measurement solution for feed resources.

Main Methods:

• Utilized a stereo camera system to capture RGB images of feed piles.
• Employed the U2-Net network for image segmentation to remove background interference.
• Developed a deep learning model integrating self-attention mechanisms and multiscale fusion with ResNet for feed amount estimation.

Main Results:

• The proposed method achieved a mean absolute error of 0.3487 kg and a root-mean-squared error of 0.4456 kg for feed amounts between 0 and 10 kg.
• Demonstrated superior performance compared to commonly used methods in real-world ranch scenarios.

Conclusions:

• The developed deep learning approach provides a highly accurate and efficient method for estimating dairy cow feed amounts.
• This technology offers a valuable intelligent measurement solution for optimizing dairy farm management and resource utilization.