The presence of weed plants in lawns disrupts their behavior and correct growth. Moreover, it implies a lack of uniformity, which is one of the most important factors of the law ns. The early detection of weeds is crucial to minimize the need for phytosanitary products. Image processing techniques and machine vision are widely used in many different areas such as agriculture, industry, or object identification. In this paper, we propose the use of image processing techniques to detect undesired grass species in the lawn. We utilize a drone with an Arduino module to take pictures. The obtained images are used to determine the best option to detect the presence of weeds. Pictures from different grass species with and without undesired weed species are used. The Red, Green and Blue (RGB) layers of each picture are mathematically combined in order to obtain a new raster layer to automatically detect the weed. Two different methods are used. Different equations offer different results depending on the weed species. We can detect two big groups of weeds with the first or with the second method, according to their color. Finally, the proposed formulas are verified with pictures taken with different solar conditions. An aggrupation method to minimize the false positives is shown.
展开▼