Methods and techniques for plant and weed detection creating a database for future computer vision systems in weed control and practical implementations: Insights from the KIdetect project, funded by the BMEL

Noori, FaryalHopf, LucasFlierl, PhilippZimmermann, AlexanderNiedermeier, MichaelHolst, GerhardSchmailzl, AntonKlein, MaikeKrupka, DanielWinter, CorneliaGergeleit, MartinMartin, Ludger2024-10-212024-10-212024978-3-88579-746-3https://dl.gi.de/handle/20.500.12116/45086This paper explores weed detection methodologies in vertical farming systems using Short Wave Infrared (SWIR) and Visual (VIS) camera technology, alongside computer vision techniques and Artificial Intelligence (AI). It investigates pixel-matching techniques for stereo-image processing to enhance imaging accuracy and reliability in agriculture. Classical methods like SIFT FLANN Matcher, Epiline Matcher, and Partial Epiline Matcher are evaluated. The paper also examines the integration of AI with classical pixel-matching methods to streamline pixel pair identification. Real-world accuracy assessments demonstrate promising results, facilitating practical applications. Additionally, it covers camera calibration and image rectification tasks on VIS cameras to support 3D reconstruction for plant structure analysis, alongside Stereo pixel matching. Overall, it provides valuable insights into stereo image analysis in agriculture, fostering future research and practical implementations in precision agriculture and computer vision systems.enVertical-Farming-SystemSWIR-TechnologyAI-weed detectionImage data generationLaboratory environment3D reconstructionAI-basedCNN disparity mapMachine LearningMethods and techniques for plant and weed detection creating a database for future computer vision systems in weed control and practical implementations: Insights from the KIdetect project, funded by the BMELText/Conference Paper10.18420/inf2024_1141617-5468