
Agricultural Image Repository
We’re building a cloud-based bank of tens of thousands of high-resolution images of cash crops, weeds, and cover crops, plus metadata that describe them. We’re using this Agricultural Image Repository to build sensors and tools to help farmers make real-time, data-driven management decisions to maximize yield and better respond to weeds, insects, and diseases.
We developed BenchBot for populating the Agricultural Image Repository. It’s an automated, high-throughput, relatively low-cost modular system that collects high-resolution plant images quickly and reliably, indoors and out.
Composed of open-source hardware and software, BenchBot has a mobile, robotic, aluminum frame that holds cameras, lighting, sensors, and computer equipment and is positioned over the growing area. It moves autonomously down a row of plants, collecting images and other data and sending them to the Agricultural Image Repository’s computer vision pipeline, where they’re processed and annotated semi-automatically.
We’re currently using BenchBot to collect thousands of images of cash crops, cover crops, and weeds in a semi-field environment (potted plants kept outdoors) from seedling stage to approximately 15.7 inches (40 cm) tall.
Access our BenchBot documentation at Confluence and our code at GitHub.
Access our Agricultural Image Repository documentation at Confluence.
A student under Lirong Xiang has developed this virtual environment for planning future uses of the BenchBot using Unreal Engine, a video game engine developed by Epic Games.
In this instance, the model BenchBot travels the X, Y, and Z axes capturing images of potted plants in a semi-field application.