Improved Real-Time Weed Detection and Assessment Framework Using an Enhanced Yolov8n Model

Weed detection and management remain major challenges in modern agriculture, as weeds The identification and control of weeds in agriculture has been a major issue to modern agriculture, since the weeds compete with crops over crucial resources, and lower yield in case they are not well controlled. Manual weeding, manual application of herbicides are traditional techniques that are expensive but unsustainable to the environment. This research experiment suggests a new design with a real-time weed detector and assessor, based on a revised version of You Only Look Once 8 nano (YOLOv8n) model, trained with a Dual Attention Network (DAN) as the backbone and a refined Feature Pyramid Network (FPN) as the neck. An ensemble dataset was generated based on 10,000 images of rice farms sampled in Ihuokpara, Enugu, Nigeria and was augmented with publicly available datasets of Roboflow. Improved YOLOv8n was trained to detect various species of weeds in diverse conditions and combined with a Weed Severity Index (WSI) algorithm to measure the growth of weed in respect with the farm area. The experimental data proved that the improved YOLOv8n model had 96% detection probability, 100% accuracy, 99% recall, and F1-score of 92%, which are much better than the traditional models. It was found that the DAN allowed better spatial and channel attention and the enhanced FPN allowed good multi-scale feature extraction, which guaranteed a good detection even in case of occlusion and variable weed sizes. The WSI model categorized severity of weed into low, moderate, and high levels of severity as an intervention to take action. The suggested system was also confirmed by means of the software implementation and is expected to be deployed on Unmanned Aerial Vehicles (UAVs) having Simple Mail Transfer Protocol (SMTP)-capable farmer notifications. The framework further develops the field of precision agriculture through deep learning, transfer learning, and automated severity assessment, thereby lowering the number of herbicides applied to the crop, increasing its yield, and enhancing the environmental sustainability of farm management methods.