Weed infestations have the potential to cause significant economic losses for farmers as a result of diminished crop yields and escalated labor and input costs linked with weed management. Traditional weed control methods often entail indiscriminate application of herbicides across entire fields, irrespective of weed density or species composition. In this investigation, we propose a hierarchical detection algorithm for identifying multi-species weeds and have devised a variable spraying system in agricultural settings. The weed classification detection algorithm comprises two components: crop seedling target detection and weed detection. The enhanced YOLOv5 algorithm was initially merged with Vision Transformer (ViT) to introduce the W-YOLOv5 crop seedling target detection algorithm. Experimental validation results indicate that the mean average precision (mAP) of the proposed W-YOLOv5 stands at 87.6%, representing a 3.2% increase over the original YOLOv5. Compared with YOLOv7, this method demonstrates a 4.4% improvement in mAP, accompanied by an 80.14% reduction in floating point operations (FLOPs). These findings underscore the effectiveness of the proposed crop seedling target detection algorithm in ensuring detection accuracy while minimizing model FLOPs. Subsequently, following the detection of crop seedlings, the Hue, Saturation, and Value (HSV) color space filtering algorithm was employed to identify the locations of all weeds in the image, thereby facilitating the detection of weeds among wheat (Triticum aestivum L.), radish (Raphanus sativus L.), cucumber (Cucumis sativus L.), soybean (Glycine max (L.) Merr.), and corn (Zea mays L.) seedlings. Finally, the severity of weed infestation in farmland was classified into five levels. Leveraging these severity levels, we developed a variable spraying system and integrated the developed algorithm into the system. Field validation experiments demonstrate that at a speed of 4 km/h, the spraying accuracy of the system can reach 90.32%, effectively enabling precise variable spraying.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call