A pest and disease classification method and system based on quaternion coding combined with a Mamba model

Abstract

The application discloses a kind of pest classification method and system based on quaternion coding and Mamba model combination, it is related to pest identification technical field, the method includes: obtaining pest image and pretreatment;For each pixel point in the image after pretreatment Structure quaternion representation;In quaternion field, structured enhancement operation is performed to quaternion representation, including the normalization processing of the module length of quaternion representation and the constrained rotation disturbance applied to quaternion representation;The imaginary part of the quaternion representation after structured enhancement operation is coded into a standard RGB format image;In the training phase, information retention reconstruction constraint is applied to the coding process;The standard RGB format image obtained by coding is input into the visual classification network based on Mamba model for sequential feature extraction and fusion;Based on the output of visual classification network, the category of pest is determined.The application significantly improves the cross-scene robustness, generalization ability, accuracy and inference efficiency of pest classification.

Images