Hyperspectral image classification method deployable by unmanned aerial vehicle

By employing a spatial spectral triangular connection structure and feature fusion method, the problem of high computational complexity in hyperspectral classification models is solved, enabling efficient hyperspectral image classification on UAV edge devices and supporting the real-time application of UAV-borne hyperspectral imaging systems in the agricultural field.

CN122244530APending Publication Date: 2026-06-19TAIYUAN INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
TAIYUAN INST OF TECH
Filing Date
2026-03-25
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing hyperspectral classification models have high computational complexity, resulting in high consumption of computing resources and making them difficult to process and deploy in real time on UAV edge devices.

Method used

By employing a spatial spectral triangular connection structure, the initial features are mapped to spatial spectral features through the first and second convolutional paths. Combined with temporal and frequency domain feature extraction, dimensionality reduction, and nonlinear activation, efficient feature fusion and classification are achieved.

Benefits of technology

While reducing computational complexity and memory footprint, it achieves efficient extraction and accurate classification of deep spatial spectral joint features of hyperspectral images, supporting real-time processing on UAV edge devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122244530A_ABST
    Figure CN122244530A_ABST
Patent Text Reader

Abstract

This invention provides a hyperspectral image classification method deployable on UAVs, comprising: preprocessing the acquired hyperspectral image to obtain initial features; performing parallel spatial spectral feature mapping on the initial features through a first convolution path and a second convolution path to obtain respective first output features and second output features; performing nonlinear activation on the second output features to obtain activation features, and fusing the activation features with the first output features to obtain local spatial spectral joint features; extracting temporal and frequency domain features from the local spatial spectral joint features to obtain temporal and frequency domain features; fusing the temporal and frequency domain features to obtain comprehensive discriminative features; and performing classification based on the comprehensive discriminative features to output the classification result of the hyperspectral image. This method reduces computational complexity and memory usage, and solves the problem that hyperspectral classification methods are difficult to deploy in real time on edge computing devices such as UAVs.
Need to check novelty before this filing date? Find Prior Art