TY - GEN
T1 - An automatic data acquisition system for acquiring training data for a deep learning algorithm for individual cow intake prediction
AU - Bezen, R.
AU - Edan, Y.
AU - Halachmi, I.
N1 - Publisher Copyright:
© Precision Livestock Farming 2019 - Papers Presented at the 9th European Conference on Precision Livestock Farming, ECPLF 2019. All rights reserved.
PY - 2019/1/1
Y1 - 2019/1/1
N2 - Individual feed intake of dairy cows is an important, currently unavailable, variable in commercial dairies. Earlier systems developed were either costly or unreliable enough for commercial farms. This research developed a low-cost individual feed intake system using RGB-D cameras and deep learning algorithm. Depth and colour images are produced from an RGB-D camera, and are used to build a CNNs (Convolutional Neural Networks) regression model for weight intake prediction. To provide training data, an automatic data acquisition system was designed to collect a wide range of food weights, in different configurations and conditions (indoor, outdoor, direct-sun). The system included a scale and a micro-controller set in the Volcani research dairy facility, an open cowshed with Holstein cows, eating Total Mix Ration. With this setup, 28,761 data were collected over seven days. Additional data were created by data augmentation methods. The model was evaluated on a test-dataset acquired in the same dairy farm. The model was tested for different combinations of training data (direct-sun/outdoor) to evaluate the importance of the data diversity. Per meal, mean absolute and square errors were 0.127 kg, and 0.034 kg2, respectively, the consumed amount of feed measured in range of 0-8 kg. The sensitivity analysis shows that the amount and diversity of data is important for model training. Better results were achieved for the model that was trained with high diversity data. The results suggest that cameras and CNNs are feasible for individual feed intake measurement on the dairy farm.
AB - Individual feed intake of dairy cows is an important, currently unavailable, variable in commercial dairies. Earlier systems developed were either costly or unreliable enough for commercial farms. This research developed a low-cost individual feed intake system using RGB-D cameras and deep learning algorithm. Depth and colour images are produced from an RGB-D camera, and are used to build a CNNs (Convolutional Neural Networks) regression model for weight intake prediction. To provide training data, an automatic data acquisition system was designed to collect a wide range of food weights, in different configurations and conditions (indoor, outdoor, direct-sun). The system included a scale and a micro-controller set in the Volcani research dairy facility, an open cowshed with Holstein cows, eating Total Mix Ration. With this setup, 28,761 data were collected over seven days. Additional data were created by data augmentation methods. The model was evaluated on a test-dataset acquired in the same dairy farm. The model was tested for different combinations of training data (direct-sun/outdoor) to evaluate the importance of the data diversity. Per meal, mean absolute and square errors were 0.127 kg, and 0.034 kg2, respectively, the consumed amount of feed measured in range of 0-8 kg. The sensitivity analysis shows that the amount and diversity of data is important for model training. Better results were achieved for the model that was trained with high diversity data. The results suggest that cameras and CNNs are feasible for individual feed intake measurement on the dairy farm.
KW - 3D camera
KW - Deep learning
KW - Individual cow feed intake
KW - Machine vision
KW - Precision livestock farming (PLF)
UR - http://www.scopus.com/inward/record.url?scp=85073762083&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85073762083
T3 - Precision Livestock Farming 2019 - Papers Presented at the 9th European Conference on Precision Livestock Farming, ECPLF 2019
SP - 284
EP - 291
BT - Precision Livestock Farming 2019 - Papers Presented at the 9th European Conference on Precision Livestock Farming, ECPLF 2019
A2 - O'Brien, Bernadette
A2 - Hennessy, Deirdre
A2 - Shalloo, Laurence
PB - Organising Committee of the 9th European Conference on Precision Livestock Farming (ECPLF), Teagasc, Animal and Grassland Research and Innovation Centre
T2 - 9th European Conference on Precision Livestock Farming, ECPLF 2019
Y2 - 26 August 2019 through 29 August 2019
ER -