Based on the increasingly severe reality of land desertification, it is of great practical significance to improve the efficiency of vegetation restoration, and technologies such as artificial intelligence, automation, and information communication will greatly improve the efficiency of vegetation restoration. In this study, a unified platform was constructed to test four activation functions (ReLU, SiLU, GELU, and Mish) under the same conditions, based on research on different neural network functions and their application in the YOLOv5 model. Through analysis of the experimental results, it was found that the YOLOv5 model with the SiLU activation function performed well in plant disease recognition tasks and showed outstanding performance in multi-target recognition.