Leveraging deep learning for dollar spot detection and quantification in turfgrass

Abstract

This study evaluates the effectiveness of fine‐tuning a semantic segmentation model to identify and quantify dollar spot in turfgrasses, the most extensively managed and researched disease of turfgrasses worldwide. Using the DeepLabV3+ model, recognized for its capability to segment complex shapes and integrate multi‐scale contextual information, the research leveraged a diverse dataset comprising various turfgrass species, disease stages, and lighting conditions to ensure robust model training. The trained model is able to identify and segment disease instances accurately and precisely, and the results indicate the potential for model‐based assessment to outperform traditional visual assessment methods in speed, accuracy, and consistency. The development of deep learning models on extensive datasets like ImageNet requires significant computational resources. However, by fine‐tuning a pretrained semantic segmentation model, we adapted it for disease segmentation using only a standard personal computer's graphics processing unit. This approach not only conserves resources but also highlights the practicality of deploying advanced deep learning applications in turfgrass pathology with limited computational capacity. The proposed model provides a new tool for turfgrass researchers and professionals to rapidly and accurately quantify this important disease under real‐world growing conditions. Additionally, the findings suggest the potential to apply deep learning algorithms to other turfgrass diseases to support data‐driven decisions. This could enhance disease management practices and improve decision‐making processes for fungicidal treatments, thereby improving the economic and environmental sustainability of turfgrass management.