Precision Agriculture: Weed and Crop Detection in Real-Time Using an Optimized YOLOv8s Model

With the application of smart agriculture, in order to solve the problems of crop yield loss, grain yield reduction and economic loss caused by weeds, the level of automation of agricultural production needs to be increased. This study focuses on three common weeds-thistle, sedge, and li-found near crops such as corn and grassland early ricegrass. These weeds are selected as research objects, and the YOLOv8s algorithm's technology for intelligent crop and weed recognition is applied. Firstly, based on crop and weed datasets, the training set and validation set are divided into an 8:2 ratio. In this paper, images of weeds in different growth stages and environments were extensively collected to build data sets, image processing technologies such as gray-scale and normalization were used for preprocessing, deep convolutional neural networks were used to learn weed features to achieve accurate species identification, and weed locations were identified through object detection techniques. Experiments demonstrate that the YOLOv8s-based farmland weed detection model exhibits excellent performance, high accuracy and high efficiency. The results are of great significance for optimizing farmland weed management, improving crop yield and quality, reducing pesticide use and promoting precision agriculture, and provide efficient intelligent technology approaches and decision support for weed control.