JoWUA

Abstract

Due to uneven spacing of the plants, identification of weeds and bushes in crops and vegetables plantations is more difficult than identification of weeds and bushes in crops. There has not been much research done on weed identification in vegetable plantations thus far. Although there is a wide variety of plant species, the traditional crop weed detection techniques are used to directly identify weeds. This research introduces a novel approach that combines image processing and deep learning techniques. Instead of directly tackling weed detection, the proposed method focuses on identifying vegetables. The process begins by utilizing a trained CenterNet model to detect and create bounding boxes around the vegetables. Subsequently, any remaining green objects outside the bounding boxes are classified as weeds. By narrowing the scope to vegetable detection, the model avoids the complexities associated with different weed species. Moreover, this approach offers the advantage of reducing the complexity of weed detection and minimizing the required training image dataset, leading to improved performance and accuracy in weed identification. To separate the weeds from the background, a color index-based segmentation technique is employed using image processing methods. Using Genetic Algorithms (GAs), the employed colour index is chosen and assessed in accordance with the Bayesian classification error. The trained CenterNet model gets score 0.953 in Figure 1, a recall of 95.2%, and a precision of 95.8% during field test. In comparison to the widely used ExG index, the PI (Proposed Index) is 19R + 24G + 2B = 864 produces outstanding segmentation accuracy at significantly less cost for computation segmentation. Outcomes of this experiment show that the suggested strategy for weed identification in vegetable plantations may be used successfully on the ground.