Weed Recognition Method Based on a Lightweight YOLOv5

Weeds, as one of the important hazards that interfere with crop growth in the production process, cause huge economic losses every year. To achieve the development of smart agriculture, solving the problem of weeds plays a key role in improving crop health and increasing yields. This article focuses on identifying common weeds in farmland and automatically recognizes them by introducing the YOLOv5 model. Based on practical application scenarios, the model has decided to replace the feature extraction network with GhostNet and make lightweight improvements to reduce the computational complexity of model parameters; And introduce the CA attention mechanism module to further enhance the ability to extract location information of detection targets and improve detection accuracy. The experimental results show that the proposed model reduces the number of parameters by 24%, floating-point operations by 32%, model size by 3.2M, and mAP by 0.8% compared to the original model.