Rapid detection method of UAV hyperspectral pine wood nematode disease based on decoupled risk estimation

Abstract

The invention relates to a rapid detection method of unmanned aerial vehicle hyperspectral pine wood nematode disease based on decoupling risk estimation. This method transforms the threshold determination problem into a risk estimation problem through category-decoupled risk estimation, avoiding the threshold adjustment step, and at the same time introduces a fully convolutional neural network into a single classification framework, and captures images with relatively small distances by using global spatial information. The dependency between distant pixels alleviates the "salt and pepper noise" phenomenon that often occurs in UAV image detection results, and compared with the single classification method based on image blocks, the method proposed by the present invention has a faster inference speed. This method can be used to detect pine wood nematode disease without manual intervention.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing of unmanned aerial vehicles, and in particular relates to a rapid detection method of hyperspectral pine wood nematode disease of unmanned aerial vehicles based on decoupling risk estimation. Background technique [0002] Pine wood nematode disease is known as the "cancer" of pine trees. Pine trees can show withered symptoms about 40 days after being infected with this disease, die in 2 to 3 months, and cause large-scale deforestation in 3 to 5 years. disaster. Identifying the location of dead pine trees and quickly cutting down diseased pine trees is one of the most effective ways to stop the spread of pine wood nematode. However, the method of manually monitoring the location of dead pine trees is time-consuming and labor-intensive, and it is difficult to quickly extract the location of dead pine trees in a large area. Among the remote sensing monitoring methods, UAV da...

AI Technical Summary

Problems solved by technology

[0003] At the same time, the detection of pine wood nematode dead pine trees based on UAV hyperspectral remote sensing images is also a difficult task: First, the traditional hyperspectral image detection algorithm cannot directly obtain the location of dead pine trees, and it is necessary to determine the threshold after additional , the category of the image pixel can be obtained, and the location of the dead pine tree can be obtained directly without threshold setting, which is essentially a more difficult problem of single classification of remote sensing images

Second, traditional remote sensing image detection methods are based on manual features, and the ability to identify diseased pine trees in complex scenes is limited

Third, the spatial heterogeneity and spectral variability brought about by the high spatial resolution of UAV data lead to the intensification of the phenomenon of the same object with different spectra and the same spectrum with different objects, and the detection method that only uses local spatial information leads to the existence of Obvious "salt and pepper noise"

The above problems limit the application of hyperspectral image detection algorithm in the detection of pine wood nematode disease

Images