Improving Tree Species Classification Using UAS Multispectral Images and Texture Measures

Abstract

This paper focuses on the use of ultra-high resolution Unmanned Aircraft Systems (UAS) imagery to classify tree species. Multispectral surveys were performed on a plant nursery to produce Digital Surface Models and orthophotos with ground sample distance equal to 0.01 m. Different combinations of multispectral images, multi-temporal data, and texture measures were employed to improve classification. The Grey Level Co-occurrence Matrix was used to generate texture images with different window sizes and procedures for optimal texture features and window size selection were investigated. The study evaluates how methods used in Remote Sensing could be applied on ultra-high resolution UAS images. Combinations of original and derived bands were classified with the Maximum Likelihood algorithm, and Principal Component Analysis was conducted in order to understand the correlation between bands. The study proves that the use of texture features produces a significant increase of the Overall Accuracy, whose values change from 58% to 78% or 87%, depending on components reduction. The improvement given by the introduction of texture measures is highlighted even in terms of User's and Producer's Accuracy. For classification purposes, the inclusion of texture can compensate for difficulties of performing multi-temporal surveys.

Current Practices in UAS-based Environmental Monitoring

“…RGB cameras are sufficient in many applications, such as land mapping, morphological studies, vegetation classification, and river surveys. More recently, researchers have also turned to modified off-the-shelf RGB and near-infrared (NIR) cameras, in pursuit of more accurate vegetation mapping [34][35][36][37][38][39]. This provides an advantage for vegetation mapping, leading to an increase in the classification accuracy of up to 15% [40].…”

“…Obtaining ground truth data is also achieved by identifying major terrain and vegetation units within the surveyed zone [88], or by selecting samples directly on images on the basis of GNSS surveyed points and handwritten documentation [18,36]. For vegetation patches that are easily distinguished from surroundings by their size, specific growth pattern or phenology traits, there is no need for detailed georeferencing.…”

Acquisition of Forest Attributes for Decision Support at the Forest Enterprise Level Using Remote-Sensing Techniques—A Review

“…They tested two different algorithms: k-nearest neighbor method (k-NN), combined with a genetic algorithm and a Random Forest method and achieved accuracy of classification 0.823 for tree species and 0.869 for tree genus. Multispectral, textural, and in addition multi-temporal data were deployed for tree species classification in Gini et al [184], Maximum Likelihood algorithm and principal component analysis were applied on individual bands, and the authors concluded that texture features produced and increase in accuracy when the values changed from 58% to 78% or 87%. Very high (or sometimes referred to as ultra-high) resolution from UAVs brings problems related to shadows and related error in classification.…”

Resnet-Based Tree Species Classification Using Uav Images

“…Thus, many researches have focused on the use of UAV imagery for tree species classification. For example, Gini et al (2018) investigated if the use of texture features, derived from UAV multispectral imagery, can improve the accuracy of tree species classification. Franklin and Ahmed (2018) have used UAV multispectral images acquired over forests to successfully classify few tree species by means of a machine-learning classifier which was found effective in separating individual crowns with spectral response, textural, and crown shape variables.…”