基于实例分割和机器学习的育肥猪群体体重估测方法研究

打开文本图片集

中图分类号：S828 文献标识码：A 文章编号：2095-5553（2025）10-0153-08

Abstract：Targeting thelarge-scale production and breeding offatening pigs，traditional pig weighing methods have problems such aslowautomation level，low eficiency，time-consumingand laborintensive，andeasy stresson pigs.A non-contact fatening pig population weight estimation method based on instance segmentation and machine learning is proposed.Two diferent instance segmentation algorithms，Mask R—CNNand Mask2former，were used to obtain the mask contours of pigs，and the results were compared. The segmentation accuracy of Mask R—CNN and Mask2former were 93.86% and 98.98% ，respectively.Finally，the Mask2former instance segmentation model was selected.Combining the image informationof thepig segmentation mask，the relevant feature parametersof themask image were extractedas thedata inputfor the model.Diferent algorithms wereused toconstruct multipleweight estimation models forcomparison.The random forest estimation model is found to have the best performance，with acoeficientof determination R2 of 0.94，an average absolute error of 7.92kg ，and an averagerelative error of 2.58% . The experiment demonstrates that the non-contact Weightestimation method forfatening pig populations based oninstance segmentationand machine learningcan efectively predict body weight，providing technical and theoretical support for achieving automatic weighing of pig populations.

Keywords：fattening pig population；weight estimation；image procesing；instance segmentation；machine learning; feature extraction

0 引言

体重是动物关键生长指标之一，其变化趋势能够反映动物的健康状况[1]。（剩余12669字）