Adapting a UAV-Based Weed Detection Method (SWIM) for a Tractor-Mounted Optical Sensor

Abstract

Site-specific weed management can reduce herbicide use and improve the sustainability of crop production, but its adoption remains constrained by the complexity and costs of many sensing technologies. Integrating weed detection algorithms into optical sensing systems already used for precision agriculture may represent a practical strategy to overcome these limitations. This study aimed to adapt the SWIM Weed Detection (SWIM-WD) module, originally developed for UAV imagery, to images acquired by the Augmenta Field Analyzer, a tractor-mounted optical sensor. The method was evaluated using images acquired in a commercial maize field. Calibration images acquired under weed-free conditions were used to derive crop structural parameters, whereas validation images containing weeds were used to evaluate two approaches, the Simple Method (SM) and an Improved Method (IM), the latter incorporating image-specific maize row detection. Both approaches provided Weed Green Cover (WGC) estimates showing high agreement with reference values (CCC = 0.87–0.92; R2 = 0.82–0.89). The findings provide preliminary evidence that SWIM-WD can be adapted to a tractor-mounted sensing platform, supporting precision agriculture systems with weed detection capabilities without dedicated hardware. Further validation across fields, crop stages, and environmental conditions is required to assess the robustness and general applicability of the method.