상세 보기
Generative AI-enhanced multimodal deep learning for pest and disease detection in Chinese herbal medicine
- Chang, An;
- Rani, Shalli;
- Kim, Byung-Gyu;
- Gong, Jianing;
- Qin, Zhiyong;
- 외 2명
WEB OF SCIENCE
0SCOPUS
0초록
Chinese herbal medicine cultivation faces mounting pressures from pest infestations and disease outbreaks that compromise both yield quality and therapeutic efficacy. Traditional monitoring approaches struggle with the vast diversity of medicinal plant species and the subtle symptom variations across different growing environments. We present GAI-MDPS, a framework that integrates generative artificial intelligence with multimodal deep learning to enable accurate disease detection from limited training samples. The architecture employs vision transformers initialized with domain-specific disease attention bias priors for extracting global visual features, while leveraging pre-trained language models to generate contextual descriptions that guide visual attention toward diagnostically relevant image regions. Cross-modal contrastive learning aligns linguistic and visual representations, creating synergistic interactions where language understanding enhances visual interpretation. Comprehensive evaluation across PlantVillage and a self-constructed TCM-Disease dataset demonstrates substantial improvements over seven baseline methods, achieving 95.3% and 89.4% accuracy, respectively, with five-fold and ten-fold cross-validation confirming result stability. The generative module produces synthetic disease images with an overall FID of 23.6, and per-severity analysis shows a 9.1 %age point advantage over the best baseline at severity level 1, where early-stage detection delivers the highest practical value for timely intervention. The framework converges 2.1 times faster than existing approaches while maintaining superior generalization across diverse cultivation environments. Ablation studies reveal that multimodal integration reduces attention dispersion by 43.2% and improves lesion localization precision, enabling reliable automated monitoring for smart agriculture platforms serving Chinese herbal medicine production.
키워드
- 제목
- Generative AI-enhanced multimodal deep learning for pest and disease detection in Chinese herbal medicine
- 저자
- Chang, An; Rani, Shalli; Kim, Byung-Gyu; Gong, Jianing; Qin, Zhiyong; Li, Keqin; Lv, Jianhui
- 발행일
- 2027-01
- 유형
- Article
- 권
- 204
- 호
- Part A
- 언어
- ENG
- 출판사
- Elsevier B.V.
- 발행국가
- 네덜란드
- ISSN
- E 1872-9681
P 1568-4946