Researchers develop learning approach for drone-based surveillance, scouting 

Unmanned Aerial Systems (UAS) aid farmers in crop disease management, field scouting and pesticide applications. UAS has the unique ability to gather aerial imagery and other large amounts of data although interpreting aerial images can be complex and time consuming.

Researchers at Purdue University in Indiana have developed AgSemiSeg, a semi-supervised learning approach designed for agricultural semantic segmentation. The objective is to have an automated field scouting system, ensuring precise identification and management of anomalies in agricultural fields real-time.

AgSemiSeg is designed to handle varying lighting conditions, occlusions due to overlapping foliage and the inherent variability in crop appearances. The semi-supervised learning methodology of AgSemiSeg helps with the usage of both labeled and unlabeled data, enhancing the model’s learning capacity while minimizing manual annotation. This will help improve accuracy and field efficiency of field image segmentation and at the same time pave the way for more advanced analytics in agricultural monitoring and management.

Agricultural companies and farmers will benefit from this advanced image interpretation model, particularly as it relates to detecting plant diseases through semantic segmentation coupled with drone imagery. By analyzing high-resolution images captured by UAVs, the model can detect the early stages of a disease. This allows farmers to take immediate action before the disease spreads to the entire crop. This saves farmers from the devastating economic losses of losing an entire crop, reduces the need for widespread pesticide application, promotes healthier crops and resilient land use.

View the full statement on the NIDB.

Project supported by Hatch funds.