Fuente: WIPO "tobacco agriculture" The invention discloses a baking process optimization method based on machine vision time sequence analysis and natural language processing. The method comprises the following steps: S1, collecting and calibrating historical tobacco leaf baking pictures; s2, training a tobacco leaf baking degree classification model; s3, collecting and discretizing a historical water loss rate; s4, discretized water loss rate data and classified tobacco leaf pictures are calibrated; s5, key point recognition model training; s6, historical expert opinions and baking curve adjustment strategies are collected and calibrated; and S7, training an expert opinion conversion model. By the adoption of the method, direct and accurate matching of expert opinions and the two parameters can be achieved, targeted improvement and optimization opinions of experts can be directly obtained, and reliable and accurate optimization of the process is achieved.