Frame-based reflectance estimation from multispectral images for weed identification in varying illumination conditions

Anis Amziane (1) , Olivier Losson (1) , Benjamin Mathon (1) , Aurelien Dumenil , Ludovic Macaire (1)

1 CRIStAL - Centre de Recherche en Informatique, Signal et Automatique de Lille - UMR 9189

To prevent the growth of weeds in precision farming, multispectral imaging has gained much interest for its ability to provide vegetation images with a high spectral resolution. However, spectral reflectance computation is an issue when the image is assembled from successive frames acquired under varying illumination conditions. In this study, we present a method to estimate reflectance from images acquired by a linescan camera in such conditions. Because rows in a given channel are associated to different illumination conditions, we process the image row-wise to improve reflectance estimation. Experimental segmentation results show that our method is a good candidate to effectively identify crops from weeds.

Mots clés: Multispectral imaging; Reflectance estimation; Illumination; Precision farming; Weed detection; Linescan camera

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Ingénierie de l'environnement; Sciences et techniques de l'agriculture

Anis Amziane, Olivier Losson, Benjamin Mathon, Aurelien Dumenil, Ludovic Macaire. Frame-based reflectance estimation from multispectral images for weed identification in varying illumination conditions. Tenth International Conference on Image Processing Theory, Tools and Applications (IPTA 2020), IEEE, Nov 2020, Paris, France. ⟨10.1109/IPTA50016.2020.9286692⟩. ⟨hal-02994446⟩