Fecha de publicación: 05/03/2024 Fuente: WIPO "tobacco agriculture" The invention provides a method for accurately identifying a damaged tobacco, which comprises the following steps of: extracting leaf color features and texture features of a tobacco leaf image through an RGB (Red, Green, Blue) model and an HSV (Hue, Saturation and Value) model; the texture features comprise texture inertia, texture energy, texture entropy and texture correlation; according to the method, more accurate feature information is provided for model construction in the next step, a machine learning-visual self-supervision pre-training model is constructed to carry out accurate classification on tobacco images, the model can rapidly converge, the recognition precision is improved, the semantic complementation ability of picture representation learning is enhanced, the model is subjected to accuracy evaluation, and the accuracy of the model is improved. A foundation is laid for providing reason analysis and prevention measures for different types of flue-cured tobacco, and reference is provided for assisting tobacco growers in flue-curing and reducing the flue-curing loss rate of the tobacco leaves.