Thermal Image Thresholding for Automatic Detection of Bovine Mastitis

Disruptive technologies are of great relevance for the advancement of animal science. Dairy cattle farming is of uttermost importance in the global scenario, but hindrances such as mastitis have been causing losses in the sector. This study aimed to develop and validate an algorithm for thresholdbased segmentation of thermal images and automatic classification of clinical cases of bovine mastitis. The research was conducted at two milk production units located in the municipalities of Capoeiras and Pesqueira, Pernambuco, Brazil. The number of samples was determined according to the selection criteria and totaled 120 animals (40 healthy animals, 40 animals with subclinical mastitis, and 40 with clinical mastitis). Thermal images of the animals' udder were made with an infrared thermographic camera; the shots were taken from the front left, front right, rear and lower mammary quarters, with four images per animal, totaling 960 thermal images for analysis of said quarters. For automatic segmentation of images, an algorithm in C++ language was developed using the OpenCV library to identify the temperature referring to each pixel on the image through interpolation by the technique of pixel thresholding and quantification in the analyzed interval (t to tmax) of each image . Healthy animals presented 11,034 to 21,521 pixels. Animals with subclinical mastitis presented 12,582 to 40,032 pixels. In those with clinical mastitis, pixels ranged from 45.223 to 49.451. The algorithm for automatic segmentation allowed distinguishing the images of healthy animals from those of animals with subclinical and clinical mastitis. The routine implemented for determining the temperature of each pixel on the image was coherent, derived from results obtained through Flir Tools software.