DIGITIZED ULTRASONIC IMAGES ON LIVE PIGS FOR PREDICTION OF MEAT PROPORTION IN THE BELLY

The aim of this study was to predict the meat proportion in the belly from digitized ultrasonic images on live pigs. 100 pigs of various gen otypes, 52 of them Pietrain, were scanned at the belly under the secon d lumbar vertebra. The pictures were taken twice, digitized into 64 gr ay-levels and saved. After a filtering process various image sections were chosen and a texture analysis was applied. Cooccurrence matrices were constructed to show the distribution of gray-levels and their int errelationships. With an iterative procedure 15 texture parameters wer e first derived from a total of 6820. Then a stepwise regression was a pplied to find equations predicting the meat proportion. The repeated pictures were sampled to a first and a second data set in order to get robust prediction equations. The equations were developed at the firs t data set and tested at the second. For the best, regression equation s applied to the second data set, multiple correlations to the meat pr oportion were 0.60 and 0.54 for all animals and for Pietrain, respecti vely. Regarding the type of analysis and further possible technical im provement, this is considered to be the lower limit of accuracy. It is expected that digital imaging of the belly may be a proper addition t o ultrasonic backfat measurements for practical breeding strategies. T his would not only mean direct selection for the quality of the belly. In the long run an indirect effect on the whole carcass quality might also be expected if selection for backfat thickness is hampered by fu rther reduction of variance due to selection.