Enabling agriculture 4.0 on analog pointer-needle setups with computer vision-based reading methods: the case for legacy penetrometer gauge reading

Abstract

Agriculture 4.0 faces significant challenges in integrating legacy analogue equipment, such as soil penetrometers, into digital workflows due to technical incompatibility and cost constraints. This study addresses these by developing a low-cost, open-source computer vision (CV) based method for automated reading of analogue penetrometer pointer meters. The methodology employs a smartphone-based image acquisition protocol with a Polylactic acid (PLA) 3D-printed mount for consistent framing and image stability. It utilizes the Segment Anything Model (SAM) to segment the gauge dial and needle. Key innovations include automated localization of dial features, angular optimization for center alignment, and needle tip alignment via iterative line sweeps. The algorithm was calibrated and validated on the PEN3-v1 dataset (3,186 images under heterogeneous lighting), achieving the following accuracy metrics: concordance correlation coefficient (CCC) of 0.9982, mean absolute error (MAE) of 0.519°, and mean relative error (MRE) of 1.545%. These results surpass volunteer human readers (MAE: 1.332°, MRE: 12.285%) and prior CV-based reading methods. The method enables precise and low-cost digitization of penetration resistance readings, reducing training required for sampling, and facilitating soil monitoring for precision agriculture. It demonstrates strong generalization potential for other analog gauges (e.g., tensiometers or pressure meters) and is fully replicable using open-source tools.