Construction of sheep forage intake estimation models based on sound analysis. (April 2020)

• Record Type: Journal Article
• Authors: Sheng, Hang; Zhang, Shengfu; Zuo, Lishi; Duan, Guanghui; Zhang, Hailin; Okinda, Cedric; Shen, Mingxia; Chen, Kailing; Lu, Mingzhou; Norton, Tomas
• Is Part Of: Biosystems engineering. Volume 192(2020)
• Page Start: 144
• Page End: 158
• Publication Date: 2020-04
• Subjects: Sheep behaviour -- Forage intake estimation -- Acoustic analysis -- MFCC -- Linear regression -- Elastic network; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2020.01.024
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : Forage intake is one of the most important indicators of health and productivity of ruminant livestock, such as sheep. Forage intake estimation can also make an important contribution to the design and implementation of rotational grazing systems. In this paper, sheep forage intake estimation models were developed based on acoustic analysis from data gathered by an audio recorder mounted on a sheep. A data set with 114 pieces of audio was constructed by collecting ingestion sound of eight female sheep (2 years old with 50 ± 3 kg body mass). A Gaussian kernel-based support vector machine (SVM) classifier was trained to identify chewing sound segments from sheep ingestion audio. Seven explanatory variables were extracted from each chewing sound segment. These variables were used to establish single variable and multiple variable-based forage estimation models. Least squares regression and elastic network approach were employed to determine the coefficients of the single variable and multiple variable-based forage intake estimation models, respectively. Validation results showed that the best single variable and multiple variable-based model could explain 71.02% and 80.94% of forage intake changes, respectively. The average accuracy of the two best models was 86.13% and 89.32%, respectively. The results suggested that an automatic system to estimate the forage intake of sheep based on a wearable audio-recorder could be developed in the future. This could contribute to sheep health disorder identification, decisions on the number of sheep for a given grassland unit, and so on. The latter is the foundation of sheep rotational grazing. Highlights: Two types of models were established to estimate sheep forage intake. Trained SVM classifier to identify chewing sound segment with an accuracy of 95.34%. Model coefficients of correlated variables were determined by using elastic network. The best model can explain 80.94% of the forage intake changes.