Fine Crop Classification Based on UAV Hyperspectral Images and Random Forest
Open Access
- 12 April 2022
- journal article
- research article
- Published by MDPI AG in ISPRS International Journal of Geo-Information
- Vol. 11 (4) , 252
Abstract
The classification of unmanned aerial vehicle hyperspectral images is of great significance in agricultural monitoring. This paper studied a fine classification method for crops based on feature transform combined with random forest (RF). Aiming at the problem of a large number of spectra and a large amount of calculation, three feature transform methods for dimensionality reduction, minimum noise fraction (MNF), independent component analysis (ICA), and principal component analysis (PCA), were studied. Then, RF was used to finely classify a variety of crops in hyperspectral images. The results showed: (1) The MNF–RF combination was the best ideal classification combination in this study. The best classification accuracies of the MNF–RF random sample set in the Longkou and Honghu areas were 97.18% and 80.43%, respectively; compared with the original image, the RF classification accuracy was improved by 6.43% and 8.81%, respectively. (2) For this study, the overall classification accuracy of RF in the two regions was positively correlated with the number of random sample points. (3) The image after feature transform was less affected by the number of sample points than the original image. The MNF transform curve of the overall RF classification accuracy in the two regions varied with the number of random sample points but was the smoothest and least affected by the number of sample points, followed by the PCA transform and ICA transform curves. The overall classification accuracies of MNF–RF in the Longkou and Honghu areas did not exceed 0.50% and 3.25%, respectively, with the fluctuation of the number of sample points. This research can provide reference for the fine classification of crops based on UAV-borne hyperspectral images.Keywords
Funding Information
- National Science and Technology Fundamental Resources Survey Project (2019FY202502)
This publication has 20 references indexed in Scilit:
- A high-performance and in-season classification system of field-level crop types using time-series Landsat data and a machine learning approachRemote Sensing of Environment, 2018
- The use of airborne hyperspectral data for tree species classification in a species-rich Central European forest areaInternational Journal of Applied Earth Observation and Geoinformation, 2016
- An assessment of the effectiveness of a random forest classifier for land-cover classificationISPRS Journal of Photogrammetry and Remote Sensing, 2012
- Vertex component analysis: a fast algorithm to unmix hyperspectral dataIEEE Transactions on Geoscience and Remote Sensing, 2005
- Mapping nonnative plants using hyperspectral imageryRemote Sensing of Environment, 2003
- Random ForestsMachine Learning, 2001
- Fast and robust fixed-point algorithms for independent component analysisIEEE Transactions on Neural Networks, 1999
- Hyperspectral image classification and dimensionality reduction: an orthogonal subspace projection approachIEEE Transactions on Geoscience and Remote Sensing, 1994
- A transformation for ordering multispectral data in terms of image quality with implications for noise removalIEEE Transactions on Geoscience and Remote Sensing, 1988
- On the mean accuracy of statistical pattern recognizersIEEE Transactions on Information Theory, 1968