Key technologies and development trends of intelligent decision-making large models for facility agriculture

Intelligent decision-making can be expected to improve protected agriculture, particularly in labor-intensive areas. Thereby, the productivity can be enhanced in facility agriculture. The intelligent decision-making technologies can also hold significant importance: One, the production efficiency and quality can be enhanced to ensure the supply, production, and income in the agricultural modernization; Another, the emerging technologies can be integrated into the precise environmental monitoring and personalized management. This research aims to focus on the multiple scenarios of the key application, such as the greenhouse environment control, modeling and prediction of the crop growth, pest and disease identification, as well as the crop phenotypic monitoring. The relevant emerging technologies were also introduced in the current protected agriculture. A systematic investigation was then made to determine the basic technologies of the visual, language, multi-modal, embodied intelligent, and large multi-agent models. The application potential of the existing large models was assessed in the key scenarios of protected agriculture. Large language models were generated from the agricultural data, providing suggestions and decision-making support to agricultural production. Crop models were established to predict the growth status of the crops in greenhouse environments. The decision-making of the large models was also utilized in the intelligent decision-making for protected agriculture. The intelligent perception of the crop semantic information was integrated with a large visual segmentation model in order to improve the accuracy and efficiency of the decision-making. The data reception, processing, feedback, and decision-making were selected in the greenhouse environment control for the protected agriculture. The physical environment was interacted to achieve the real-time environmental regulation. Embodied intelligence also emphasized that the agents were suitable for the complex environments in both the digital and physical worlds. The multi-modal data of the embodied intelligence depended mainly on the architecture training of the large model for the protected agriculture. A multi-agent large model consisted of multiple interactive AI agents that operated independently to make decisions and then take actions autonomously, according to the environmental changes. The entire production cycle of protected agriculture, data integration, and sharing was achieved after data integration. Multiple large model was collaborated and then interacted for the intelligent decision-making. The multi-model collaboration balanced the advantages of each model. The information mining and accurate analysis were conducted to improve the efficiency and quality of the agricultural production. In conclusion, the large model was improved with traditional protected agriculture, such as information perception, growth model construction, and precise decision-making. An intelligent decision-making system was constructed to promote protected agriculture. The crop growth and the environmental trends were more accurately evaluated after the processing of the multi-source heterogeneous data using large models. The finding can provide a scientific and precise decision-making basis for agricultural production. Intelligent decision-making was also the key driving force for agricultural development. The prosperous agriculture was promoted to fully explore the multi-source heterogeneous data, in order to unlock the potential value of data. An intelligent decision-making system was accelerated to construct for the various application scenarios of protected agriculture using large models. The finding can also provide scientific guidance to reduce the production costs for the high economic benefits.