Constructs a large-scale dataset of 20, 943 image captions covering over 60 plant species and 300 diseases to support precise plant disease description generation. Proposes the PDPC framework, which captures and refines dependency relations in descriptive texts, integrates key image concepts, and restructures text for accurate depiction of plant disease phenotypic information. Demonstrates through extensive experiments that the proposed framework significantly outperforms existing models in describing plant disease characteristics. Agriculture is the foundation of global food security and quality of life, with staple crops such as rice, wheat, and maize meeting the dietary needs of the majority of the world's population. These crops are susceptible to diseases that can lead to significant yield losses; for example, wheat rust disease causes annual losses that exceed 2. 9 billion. Accurate captioning of the phenotypic characteristics of plant diseases plays a crucial role in supporting diagnosis, which is essential for ensuring food security. Existing methods in agriculture struggle to adequately address the heterogeneity in visual phenotypes and disease descriptions, which leads to inadequate focus on key disease characteristics. To address this issue, we propose a zero-shot image captioning framework named PDPC. PDPC employs an extensive descriptive corpus, syntactic analysis, and optimization of semantic structures to significantly improve the quality and generalization of disease descriptions. Additionally, we construct a dataset comprising 20, 943 image captions that describe the characteristics of plant diseases in more than 60 plant species and 300 diseases. Experimental results demonstrate that the PDPC framework outperforms existing models in accurately describing the characteristics of plant disease. The introduction of this innovative framework enhances the accuracy of disease descriptions and provides robust support for the intelligent diagnosis and management of plant diseases, ultimately paving the way for better plant health and higher agricultural yields.
Xie et al. (Sun,) studied this question.