Three stages of model training and experimental results
Our approach uses three stages to embed image features into semantic structures, resulting
in
descriptions that retain feature features and are more appropriate for phenotypic descriptions. The
experimental
results prove the effectiveness and superiority of our method, indicating that it has a broad application
prospect in the field of plant disease diagnosis.
Stage One and Stage Two
In the first and second stages, the model obtained the structural features of special sentences for
plant shape description by shielding part of the disease image feature description and adding
reconstruction statements to the phenotype description. The main goal of these two stages is to
optimize the alignment between image features and text descriptions so that the resulting
descriptions accurately reflect the disease characteristics of the plant.
Final Stage
In the final stage, the model is trained to obtain image-text feature selection and alignment, and
dialogue descriptions are generated through the model's visual and text feature extractors. This phase
focuses on generating high-quality conversation descriptions through visual and textual feature
extractors to provide more accurate disease diagnosis and phenotypic description.
Experimental Result
Through experimental validation, our approach outperforms multiple cutting-edge models in testing.
Compared to larger models, our approach outperforms GPT-4 and GPT-40 on multiple feature descriptors.
These results show that our approach can provide more accurate disease diagnosis and phenotypic
description while maintaining high efficiency.
