Aplicativo para detecção de anemia em ovinos

Abstract

Infections caused by Haemonchus contortus (HC) have caused significant losses in sheep farming, directly affecting herd health and reducing producer productivity. One of the most widely used methods for diagnosing anemia associated with this parasitosis is the Faffa Malan Chart (FAMACHA©), which is based on the visual assessment of the ocular mucosa coloration to classify animals into five levels. Despite its simplicity, the method presents limitations such as the subjectivity of the analysis and the dependence on prior experience by the evaluator. With advances in mobile device technologies and Deep Learning (DL) techniques, the possibility arises to automate this process, making it more accurate, standardized, and accessible. In this work, a mobile application (APP) was developed for the Android platform, using the Kotlin programming language together with the Jetpack Compose library. The proposed system performs ocular region detection, followed by classification into the five FAMACHA© levels, providing the producer with an indication of whether the animal is healthy or sick. The database contains 165 images, equally distributed among the five categories defined by FAMACHA©. For the detection stage, the YOLOv5 architecture was employed, achieving a mAP@0.5 of 0.995 on the test set with 17 samples. For classification, the best model presented a mean F1-Score of 0.9026 ± 0.0515 with five-fold cross-validation. The inference time measured was 83 milliseconds for detection and 20 milliseconds for classification, validating its use on mobile devices. In conclusion, the developed APP proved to be viable and effective as a support tool in the sanitary management of sheep, contributing to production sustainability and the rational use of anthelmintics.