Multi-modal data distillation agricultural pest detection large model construction method and system

Abstract

The invention discloses a multi-modal data distillation agricultural pest detection large model construction method and system. The method comprises the following steps: preprocessing to obtain pre-reserved data sets S1, S11 and S12; a base large model is used for distinguishing S11 into data RS1 capable of being correctly answered and data IS1 incapable of being correctly answered, a pre-reserved data set S3 is finally obtained in a large model scoring mode, a data set S1 with a long thinking chain and a data set S12 are combined to obtain a pre-reserved data set S2, LoRA fine tuning is carried out based on a honesty large model, and a data distillation large model is obtained; determining a quantization parameter and a quantization algorithm of the data distillation large model, and performing quantization perception training optimization and precision calibration to obtain a quantization large model; and respectively carrying out standardization processing and dichotomy judgment on information input by a user or images automatically captured by camera shooting to obtain a final answer. The problems that a traditional large model is not professional, difficult to deploy and illusion in the aspect of agricultural diseases and insect pests are effectively solved.

Images