Comparison of Single Image Processing Techniques and Their Combination for Detection of Weed in Lawns

Abstract

[EN] The detection of weeds in lawns is important due to the different negative effects of its presence. Those effects include a lack of uniformity and competition for the resources. If the weeds are detected early the phytosanitary treatment, which includes the use of toxic substances, will be more effective and will be applied to a smaller surface. In this paper, we propose the use of image processing techniques for weed detection in urban lawns. The proposed methodology is based on simple techniques in order to ensure that they can be applied in-situ. We propose two techniques, one of them is based on the mathematical combination of the red, green and blue bands of an image. In this case, two mathematical operations are proposed to detect the presence of weeds, according to the different colorations of plants. On the other hand, we proposed the use of edge detection techniques to differentiate the surface covered by grass from the surface covered by weeds. In this case, we compared 12 different filters and their combinations. The best results were obtained with the Laplacian filter. Moreover, we proposed to use pre-processing and post-processing operations to remove the soil and to aggregate the data with the aim of reducing the number of false positives. Finally, we compared both methods and their combination. Our results show that both methods are promising, and its combination reduces the number of false positives (0 false positives in the 4 evaluated images) ensuring the detection of all weeds.