Creación de mapas de cobertura de malas hierbas para dar soporte a estrategias agroecológicas

The increasing reliance on synthetic herbicides in agricultural systems has led to negative impacts on environmental health, biodiversity, and the sustainability of agroecosystems. This work proposes an innovative methodology for the identification and localization of weeds through the use of computer vision, image processing, and automatic classification techniques, with the aim of optimizing crop management and reducing the use of agrochemicals.

The study was conducted in a persimmon field in Valencia using an autonomous robot. From the captured images, 3D point clouds and orthomosaics were generated, enabling the application of vegetation segmentation, image enhancement, and keypoint detection techniques to isolate individual weeds. Subsequently, the regions of interest were analyzed using third-party classification tools to automatically identify plant species.

The results show that the proposed methodology enables the identification of weed species under real field conditions, even in scenarios with varying lighting and plant density. The validation of the classification results is beyond the scope of this work, as it requires verification by expert botanists. However, the developed methodology allows for the generation of accurate coverage maps that support agronomic decision-making, enabling targeted treatment application, reducing herbicide use, and promoting more sustainable agroecological practices.

Master’s Thesis submitted to the Complutense University of Madrid