A cow breeding Uruguayan entrepreneur aimed to optimize the detection of a health condition in cattle known as Lameness. This condition can greatly impact animal welfare and productivity, leading to significant economic consequences for the business.
Detecting Lameness early on is challenging, requiring alternative processes that offer reliable and consistent results. By observing deviations in the cow's back, an expert can identify and quantify the problem. The goal is to develop an Artificial Intelligence (AI) model that can automatically and precisely perform this process.
To create an MVP solution, Rootstrap had to address risks and establish a clear technological approach. This entailed exploring different paths to solve the problem and training the algorithms accordingly. Additionally, it involved identifying the hardware and software platforms necessary for implementing the solution architecture.
We researched and selected suitable algorithms for image segmentation using machine learning. Our goal was to detect and outline objects in images.
We evaluated both supervised and unsupervised options for automatic segmentation, involving intensive model training and data labeling. We utilized Mask R CNN with a labeled training dataset, COCO, to build an initial binary classifier for identifying healthy and unhealthy cows.