Agricultural product damage identification method, device and system based on hyperspectral image

The invention discloses an agricultural product damage identification method, device and system based on a hyperspectral image, and the method comprises the steps: collecting a sample image data set and dark field and white board images under each wave band, and correcting the hyperspectral image data to obtain a processed hyperspectral image data set; performing dimensionality reduction on the hyperspectral image data, inputting the hyperspectral image data into an agricultural product damage identification model, and performing training by adopting an SGD network optimization method of an attenuation learning rate to obtain a trained agricultural product damage identification model; and inputting the divided blocks into a trained agricultural product damage identification model in sequence to obtain segmented images of each divided block, and splicing the segmented images in sequence to obtain a complete damage identification result. According to the method, the damage identification efficiency is improved,