Scientists at the US Department of Agriculture's Agricultural Research Service (ARS) are building a national digital platform, the Digital Agricultural Systems Hub (DASH), that uses computer vision and artificial intelligence to help researchers and farmers see what is happening in their fields and respond faster.
DASH develops open-source training datasets, machine learning models and sensing tools to detect crop stress, identify weeds and pests and support management decisions. It combines imagery and sensor data from satellites, drones, field robots and farm equipment. "We are combining these data streams with AI to unlock and scale precision agriculture for farmers, from individual plants to whole farms," said Steven Mirsky, ARS director of digital agriculture and DASH co-director in Beltsville, Maryland.
At its core is the National Agricultural Image Repository (AgIR), an open, cloud-based library of more than 50 terabytes of labelled images of crops and weeds. Many are produced by BenchBot, a modular automated camera system built by ARS researchers to collect and annotate images quickly for training models.
The models reach the field through ModCam, a modular camera system mounted on equipment such as spray booms that maps and identifies plants in breeding trials, research plots and farmers' fields. Amanda Hulse-Kemp, DASH co-director in Raleigh, North Carolina, said the team is using computer vision for precision nitrogen management with cover crops and for weed mapping, and is working with sugar beet growers in Fargo, North Dakota, to map cercospora leaf spot in real time.
The work responds to pressures from weather variability, herbicide-resistant weeds, soil degradation, labour shortages and rising costs as fewer farmers manage larger operations. All DASH data and models run on SCINet, USDA's high-performance computing system.
The story is based on an ARS article carried by AgroPages.
Source: AgroPages
Comments
(0)