Towards automated weed detection through two-stage semantic segmentation of tobacco and weed pixels in aerial Imagery

In precision farming, weed detection is required for precise weedicide application, and the of tobacco crops necessary pesticide application on leaves. Automated accurate weeds through aerial visual cues holds promise. Precise in crop field imagery can be treated as a semantic segmentation problem. Many image processing, classical machine learning, deep learning-based approaches have been devised past, out which techniques promise better accuracies segmentation, i.e., pixel-level classification. We present new method that improves inter-class classification pixels. The technique applies two stages. stage I, binary classifier developed to segment background vegetation. II, three-class designed classify background, weeds, tobacco. output first input second stage. To test our classifier, dataset was captured manually labeled pixel-wise. two-stage architecture has shown precision. intersection over union (IOU) improved from 0.67 0.85, IOU enhanced 0.76 0.91 with approach compared traditional one-stage application. observe I shallower, smaller model enough where network more neurons serves purpose good detection.