Estimating the weight of Ongole crossbreed cattle based on image data using CNN and linear regression methods

Abstract

One of the contributors to the need for food, especially meat, is the Ongole breed of cattle or commonly known as PO cattle. In these livestock activities it is necessary to monitor the weight of the cattle with the aim of assessing the selling price of the cattle and knowing the health condition of the cattle. Currently breeders are still using traditional methods such as forecasts or scales in measuring the weight of cattle. Therefore, in this study using a camera sensor as an alternative instrument for measuring cattle weight. The stages of the research included image data acquisition, pre-processing, body segmentation of cattle, weight estimation and system evaluation. The process of acquiring image data is obtained with a DSLR camera device. Pre-processing is done using a kernel sharpening filter. Cattle body segmentation uses the Mask R-CNN method. The body image of the cow is then processed for weight estimation training using the CNN and Linear Regression methods. The system evaluation results at the segmentation stage succeeded in obtaining an Intersection over Union (IoU) metric value of 0.86. The weight estimation results managed to get a RMSE metric value of 1.10, MAE metric 0.24, MAPE metric 0.06%, and R2 metric 0.99.