On February 10, the world's first large model for the pear industry, named "Li Xiang," and the first large model for the soybean industry, named "Feng Shu," were officially launched in Hefei, Anhui. These models represent China's inaugural specialized systems deeply integrating artificial intelligence technology into the vertical sectors of pears and soybeans.
The "Li Xiang" large model encompasses eight functional modules: historical culture, germplasm resources, smart breeding, gene bank spectra, literature analysis, efficient cultivation, pest control, and storage processing. It integrates over ten thousand documents, tens of millions of words, tens of thousands of question-answer pairs, thousands of images, hundreds of thousands of phenotypic data points, and hundreds of terabytes of genomic data related to the pear industry. The model independently developed a core AI-based smart breeding model and utilizes advanced deep learning techniques and knowledge graph construction to enable intelligent data parsing and efficient application.
Wu Jun, Academic Dean of the College of Horticulture at Anhui Agricultural University, explained that the model aims to efficiently convert vast pear专业知识 into practical productivity, precisely serving the needs of researchers, producers, and consumers in acquiring pear industry knowledge, technological innovation, and application. It seeks to drive intelligent upgrades across the entire chain from "orchard to fruit plate" using artificial intelligence, addressing bottlenecks such as long breeding cycles and low efficiency, and providing innovative solutions with Chinese characteristics for the high-quality, efficient, green, and smart development of the pear industry.
The "Feng Shu" large model focuses on key bottlenecks in the soybean industry, integrating six functional modules: bean encyclopedia, bean molecules, bean literature, bean diseases, bean phenotypes, and bean breeding. It consolidates over 30,000 high-quality data entries, including core soybean germplasm, phenotypic images, and scientific texts, and constructs a knowledge graph for the soybean domain containing 20,000 entities and 100,000 relationships. The accuracy rate for data cleaning and annotation reaches 98%, with prediction accuracy for key traits such as "lodging resistance, disease resistance, and protein content" exceeding 90%.
Wang Xiaobo, Dean of the College of Agronomy at Anhui Agricultural University, stated that the model has established the first generative AI breeding platform for the entire lifecycle of soybeans, achieving precise prediction of key traits, intelligent mining of superior alleles, virtual design of optimal parental combinations, and directional creation of new germplasm. It aims to provide robust technological support for breaking through soybean yield bottlenecks, accelerating the breeding of breakthrough varieties, and strengthening national food and oil security.
Zhang Qingliang, Secretary of the Party Committee of the university, emphasized that the institution will continue to focus on the goal of building a strong agricultural nation, deeply promoting the integration of artificial intelligence and agricultural technology. Taking the release of these large models as an opportunity, the university will strive to drive the transformation of agricultural research and production from traditional "experience-driven" to modern "model-driven" approaches, and endeavor to build a comprehensive smart innovation system covering the entire industry chain from "images—data—genes—germplasm—varieties—technologies—standards," contributing more wisdom and strength to accelerating agricultural and rural modernization and supporting high-quality agricultural development.
Comments