Detection of narrow-leaved weeds in chickpea based on visible remote sensing

Weed control is essential in many rainfed crops to ensure an optimal harvest. In chickpea, there are specific phytosanitary products for narrow-leaved weeds. Nonetheless, its use must be efficient, avoiding applying the products in areas without the prevalence of weeds. Besides precision agriculture allowing optimal management of phytosanitary products, a detection method is required to indicate in which areas the product should be sprayed. In this paper, we propose the use of a low-cost drone with an RGB camera to generate an index to detect the presence of narrow-leaved weeds in chickpea. RGB index has been previously proposed for broad-leaves weeds, but no index has been tested in narrow-leaves weeds. The proposed index uses the three bands of the RGB picture. The index was tested at four different relative flying heights, from 6 to 14 m. Images were obtained in a commercial exploitation, which was sown in rows. The crop was in an adult stage, before blooming. Among the identified weeds, the most common one was the Lolium rigidum. The soil is characterized by reddish tones with the presence of medium-size white rocks. The results indicate that the index can identify both narrow-leaves and broad-leaves weeds at all tested heights correctly. The lower the flying height, the more accurate the results are. The proposed index presents a series of false positives, most of them related to white rocks. This is easy to avoid through the inclusion of a secondary index. The rest of the false negatives are linked to chickpea leaves. To reduce these false negatives, the images are reclassified, and filters are applied. Among the tested filters, Laplacian 5×5 is the one with better performance for images gathered at a lower height. Nevertheless, the use of blur is preferred for images gathered at a higher height.