Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Evaluating a LiDAR sensor and artificial neural network based-model to estimate cattle live weight
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.orgCitation: 10th International Livestock Environment Symposium (ILES X) .(doi:10.13031/iles.18-004)
Authors: Rafael V. Sousa, Rubens A. Tabile, Ricardo Y. Inamasu, Luciane S. Martello
Keywords: Predictive model, Weight estimation, 3D scanner, Machine learning, Precision livestock farming.
The measurement of the animals' live weight is an important parameter for management activities with significant economic impact in animal production. Traditionally this evaluation (live weight) requires handling, often labor, and animal restraint. The objective of this paper is developing and evaluating a sensor platform to estimate animal´s live weight in feedlot finishing cattle based on an artificial neural network based-model (neural model) and using a Light Detection and Ranging (LiDAR) sensor. A sensor set consisting of a two-dimensional (2D) laser scanner and an encoder was built for the three-dimensional (3D) scanning of the back area of 107 Nellore cattle (cloud of points). The scanning data were taken in four periods every 28 days, which allowed generating 304 clouds of points after eliminating the outliers. An algorithm was constructed using “delaunay” and “convex hull” methods for obtaining the cattle rump height (Hr) and the area of the back view (Ad). The neural model with a feed-forward and multi-layered architecture was constructed and accepted Hr and Ad as inputs for estimating the live weight (Lw). A fine-tune process was performed with cross-validation protocols for optimizing the model parameters. The performance of the model was evaluated comparing the estimated and measured cattle weight through the linear regression parameters. The coefficient of determination was 0.85 and mean absolute percent error was 4.57% on previously unseen data. The results demonstrate the potential of the proposed platform and point to possible improvements to constitute a non-invasive and accurate tool to estimate the weight.(Download PDF) (Export to EndNotes)