Application of YOLOv8 Improved Model Incorporating Attentional Mechanism in Weed Identification

Accurate identification of weeds plays a pivotal role in precision agriculture, as it supports targeted herbicide application and enhances crop productivity. This paper presents an improved YOLOv8-based object detection framework that integrates a global attention mechanism to strengthen weed recognition performance. Although YOLOv8 demonstrates strong baseline accuracy and speed, it lacks sufficient global contextual modeling, which constrains its ability to capture fine-grained features in visually complex or dense field scenarios. To overcome this limitation, we incorporate a Global Attention Module (GAM) after each C2f block in the backbone, enabling the network to capture long-range dependencies and highlight semantically relevant regions. Experiments on a custom weed dataset indicate that the attention-enhanced model considerably surpasses the baseline YOLOv8 in precision, recall, and mean Average Precision (mAP), especially under challenging conditions with small-scale or overlapping targets. Furthermore, the proposed method attains a favorable trade-off between detection accuracy and computational efficiency, making it suitable for real-time smart farming applications.