Sistema Automatizado para Identificação de Fenótipo Relacionado a Precocidade e Fertilidade de Fêmea Bovina

Abstract

The identification of bovine precocity, reproductive characteristics are extremely important, because with the correct diagnosis it is possible to achieve an increase in productivity and consequently the profitability of the sector. However, this process is performed subjectively, which can lead to inaccuracy, and different results according to the evaluator. In order to present an automated solution with greater objectivity, this paper developed a methodology based on visual scores to identify the precocity of bovine females. The methodology consists in capturing images of the animal when it passes through the mango tree and through artificial intelligence algorithms it is possible to identify its precocity and fertility. The developed algorithms allow the input of the bovine image and return the probability of accuracy and fertility in the first reproductive season. To accomplish this work, images of the animals were collected from the Sete Estrelas Farm, which served as a basis to develop the algorithm. The performance of 5 algorithms that perform the identification and classification of each female bovine image were tested in order to indicate which technique is more suitable for the purpose. The results obtained through the algorithms were: RCNN in conjunction with U-Net, obtained a hit percentage of 71.42\%; the Efficientnet algorithm obtained 70.46\%; likewise the Twins algorithm obtained 70.46\%; and the best result obtained was through the Resnet algorithm with 74.92\%. The results obtained are promising, but it can be improved with the adequate capture of images of the animals and increasing the image bank, presenting a great potential for solving the problem.

Key words: Identification of bovine fertility. Female bovine precocity. Mask RCNN Algorithm. U-Net Algorithm. Resnet Algorithm. Twins Algorithm. Efficientnet algorithm.