Digital agriculture systems are encountering a deployment barrier: While the field is rapidly demonstrating intelligent technologies with genuinely exciting potential, it is remarkably hard to move solutions from the lab to the field. Our CIDA effort tackles an important question in imaging for dairy mastitis detection but also seeks to bridge this gap by viewing the entire "deployment stack" from the bottom up as contributors to the research question, asking what limitations arise in today’s settings and how those can be overcome to create better options for the future. As a result, our work is unusual in covering a much broader spectrum than might otherwise be considered.

We hypothesize that delay-sensitive edge intelligence frameworks can be deployed in agriculture to meet the time-sensitive and data-heavy processing needs it requires. More specifically, we aim to address the challenges of deploying fast and efficient computer vision and machine-learning techniques in these environments, using the dairy industry as a use case. Our objective is to use large volumes of video streaming data from a commercial dairy farm and develop delay-sensitive intelligence systems that can generate a mastitis risk score based on computer vision models derived from our teams’ earlier works. Machine learning models for teat shape, condition, and hyperkeratosis scores will be integrated with other co-variates, such as age and milking performance metrics.

Our target scenario was selected in part for its importance even to smaller dairy farmers, and as such, progress should assist a community often overlooked in digital agriculture efforts. But improvement would also have broader ramifications: if we can overcome the barriers in deploying early mastitis solutions, our insights will transfer to other domains that confront similar barriers. As such, our effort can advance the goals of the full CIDA initiative, both scientifically and in the social and economic arenas, by, most importantly, lowering the barriers to accessing AI technologies in the small farm setting.