Determination of live weight of pigs from dimensions measured using image analysis

Determination of the live weight of pigs from dimensions measured using an image analysis system has been studied in an experiment at the Research Center Foulum, Tjele, Denmark. The experiment included 416 crossbred pigs and compared ad libitum versus restricted feeding. Each pig was weighed and video recorded (from above) five times during the growing period from 25 to approximately 100 kg. The main purpose of the experiment was to estimate the precision of live weight prediction using image analysis. Spline functions were used to express the relationship between the body area of the pig measured by image analysis and the live weight of the pig for each of the three replicates of the experiment. Subsequently, a mixed model was used to examine the random effect of pigs within blocks. The error expressed in terms of the standard deviation of the residuals from the spline functions as percent of the mean was estimated as 8-9%, while it was estimated as 5-6% by the mixed model. The relationship between body area and live weight was almost linear on a logarithmic scale, and the residual standard deviation was constant. The relationship differed between replicates, presumably due to systematic error in the measurement technique. Clear differences in the size of the residuals were demonstrated for pigs from different breed combinations, with Danish Yorkshire having a higher weight than Danish Landrace at the same body area. The feeding methods had only minor influence on the relationship. Even though the precision is fairly low, image analysis techniques for weight determination of pigs seem to be promising, because they can be used with relatively low price standard equipment, and they give possibility of analysing numerous pigs as well as of other surveillance purposes in the herd. However, further research is needed into techniques for automatic image analysis, possibly using a three-dimensional approach, and into techniques for automatic calibration of herd-/breed-specific area-weight relationships.