| Popis: |
Biometricidentificationprovidesanimportanttoolforprecisionlivestockfarming.Thisstudyinvestigatestheeffectofweightgainandsheepmaturationonrecognitionperformance.Sheepfacialidentificationwasimplementedusingtwoconvolutionalneuralnetwork(CNN)calledFasterR-CNN,and ResNet50V2,equippedwiththestate-of-artAdditiveAngularMargin(ArcFace)lossfunction.Theidentificationmodelwastestedon47youngsheepatdifferentstages,duringa3-monthgrowthperiod,when theywerebetween2and5monthsold,throughoutwhichthesheepgainedapproximately30kilograms inweight.Resultsrevealedthatwhenthemodelwastrainedandtestedonimagesofsheepaged 2months,theaverageaccuracyofthegroupwas95.4%,comparedwith91.3%whentrainedonimages ofsheepaged2monthsbuttestedonimagesofsheepaged5months. |
