Research

AI-generated text

Researchers test cucumber-measuring vision systems on low-power hardware

Researchers at the University of Windsor and Mohamed bin Zayed University of Artificial Intelligence tested a computer-vision pipeline for measuring cucumber length on two low-power devices.

Researchers test cucumber-measuring vision systems on low-power hardware

Researchers at the University of Windsor and the Abu Dhabi-based Mohamed bin Zayed University of Artificial Intelligence have examined how well different computers can run a vision system that measures cucumber length. They compared two low-power devices, the NVIDIA Jetson Orin Nano and Raspberry Pi 4, against a workstation equipped with an RTX 3050 Ti graphics card.

The measurement pipeline accounts for the cucumber's curvature. It combines object detection using YOLO26n, prompt-based segmentation and an estimate of an arc fitted to the specimen's centreline. The researchers processed the same 50 greenhouse captures on all three hardware platforms; the images showed seven cucumbers in total. They used specimen-level cluster-bootstrap statistics to evaluate the results.

Different trade-offs

Available computing capacity matters when such measurements are put into practice. A scouting platform used to grade cucumbers can travel between rows carrying a single-board computer with a power budget of only a few watts. Such a device has no discrete graphics processor and may not have active cooling.

The study found that both the Jetson and the Raspberry Pi produced usable results, but suited different priorities. The Jetson with native masks was a suitable choice for throughput, while the Jetson using SAM was suited to accuracy. The Raspberry Pi with native masks offered a lower-cost option.

The experiment also illustrates why adapting AI systems to lower-power hardware matters. To automate tasks now carried out by a combination of people and machines, computer vision must be affordable to run within the computing and energy resources available at the site.