Prediction of liveweight and carcass of slaughtering cattle by digital images, using artificial neural networks and regression methods

Computer and internet technologies are widely used in different sectors. In this context, it is observed that the use of Digital Image Processing methods has increased rapidly all the time. Digital Image Analysis and Processing methods are used in many stages such as automatic product identification, classification, data acquisition and interpretation in plant and animal production. Recent years, the Artificial Neural Networks (ANN) method has been used very successfully in the solution, analysis of the problems and estimation of variables which are difficult or impossible to be realized by traditional programming methods. ANN can be applied in various fields such as feeding and breeding of livestock and it is still continuing to open new dimensions to scientific researches on applying into practice. In this thesis; the pre-slaughter live weight (LW) and hot carcass weight (HCW) of beef cattle were estimated by digital body measurements using ANN and Regression (REGR) methods and prediction models were developed. The data obtained in this study belong to the cattle with a total of 3433 numbers brought to the slaughterhouse were used. These animals were composed of 2472 Holstein, 79 Brown Swiss, 488 Simmental and 394 Other breeds. Digital body length (DJBL), digital wither height (DJWH), digital body depth (DJBD) and digital hip width (DJHW) were used for estimating LWs from digital images. Similarly digital carcass length (DJCL) and digital carcass width (DJCW) parameters were used for estimating HCWs. In this study, the highest correlation coefficient among the LWs with the digital body parameters was found in DJBL parameter in the data classified as "All data" and "Breeds", while the highest correlation coefficient between the digital carcass parameters and the HCWs was found to be DJCL parameter. In LW estimations of "All data", the coefficient of determination in the two-variable model using the DJBL and DJHW parameters with REGR and ANN methods, R2 = 88.1 and 94.3% respectively, and the highest R2 values in models with one-variable in ANN and REGR for DJBL parameter were R2= 82.0 and 90.8 % respectively. When we look at the LW estimations in terms of breeds, the highest coefficient of determination R2 = 93.0 % was found in Simental breed by REGR method and the highest coefficient of determination R2 = 97.8% was found in Brown Swiss breed by ANN models consisting of DJBL and DJHW as parameters. In HCW estimation of "All data", the coefficient of determination was found to be 75.3 and 88.5 % respectively, in the models with two-variables containing DJCL and DJCW parameters by REGR and ANN methods. When we look at the HCW estimations on the basis of "Breeds", the highest coefficient of determination by REGR and ANN methods was found in Brown Swiss breed with 84.4 and 93.4 % respectively in the models containing DJCL and DJCW as two parameters. On the basis of "All data" and "Breeds", the difference between the observed LWs and HCWs and the estimated weights obtained by REGR and ANN methods using digital images were not statistically significant (P> 0.05). In this thesis, different models for each breed were obtained which are commonly grown for beef production. The specific models obtained for the breeds should be tested with independent data sets and improved by investigating the accuracy levels of models. In addition, the use of the LW and HCW prediction models for each breed is considered to be beneficial for the sectors related to beef production.