UAV ACTUAL: 3D Convolutional Neural Networks

Mostrando entradas con la etiqueta 3D Convolutional Neural Networks. Mostrar todas las entradas

domingo, 15 de noviembre de 2020

Aspen detection in boreal forests: Capturing a key component of biodiversity using airborne hyperspectral, lidar, and UAV data

Importance of biodiversity is increasingly highlighted as an essential part of sustainable forest management.

As direct monitoring of biodiversity is not possible, proxy variables have been used to indicate site's species richness and quality. In boreal forests, European aspen (Populus tremula L.) is one of the most significant proxies for biodiversity.

Aspen is a keystone species, hosting a range of endangered species, hence having a high importance in maintaining forest biodiversity. Still, reliable and fine-scale spatial data on aspen occurrence remains scarce and incomprehensive. Although remote sensing-based species classification has been used for decades for the needs of forestry, commercially less significant species (e.g., aspen) have typically been excluded from the studies.

This creates a need for developing general methods for tree species classification covering also ecologically significant species. Our study area, located in Evo, Southern Finland, covers approximately 83 km2, and contains both managed and protected southern boreal forests. The main tree species in the area are Scots pine (Pinus sylvestris L.), Norway spruce (Picea abies (L.) Karst), and birch (Betula pendula and pubescens L.), with relatively sparse and scattered occurrence of aspen.

Along with a thorough field data, airborne hyperspectral and LiDAR data have been acquired from the study area. We also collected ultra high resolution UAV data with RGB and multispectral sensors. The aim is to gather fundamental data on hyperspectral and multispectral species classification, that can be utilized to produce detailed aspen data at large scale. For this, we first analyze species detection at tree-level. We test and compare different machine learning methods (Support Vector Machines, Random Forest, Gradient Boosting Machine) and deep learning methods (3D Convolutional Neural Networks), with specific emphasis on accurate and feasible aspen detection.

The results will show, how accurately aspen can be detected from the forest canopy, and which bandwidths have the largest importance for aspen. This information can be utilized for aspen detection from satellite images at large scale.

lunes, 12 de octubre de 2020

Tree Species Classification of UAV Hyperspectral and RGB Imagery with Deep Learning Convolutional Neural Networks

Interest in UAV solutions in forestry applications is growing.

Using UAVs, datasets can be captured flexibly and at high spatial and temporal resolutions when needed.

In forestry applications, fundamental tasks include the detection of individual trees, tree species classification, biomass estimation, etc. Deep Neural Networks (DNN) have shown superior results when comparing with conventional machine learning methods such as MLP (Multi-Layer Perceptron) in cases of huge input data.

The objective of this research is to investigate 3D Convolutional Neural Networks (3D-CNN) to classify three major tree species in a boreal forest: pine, spruce, and birch. The proposed 3D-CNN models were employed to classify tree species in a test site in Finland. The classifiers were trained with a dataset of 3039 manually labelled trees. Then the accuracies were assessed by employing independent datasets of 803 records.

To find the most efficient set of feature combination, it was compared the performances of 3D-CNN models trained with HS (HyperSpectral) channels, Red-Green-Blue (RGB) channels, and Canopy Height Model (CHM), separately and combined. It is demonstrated that the proposed 3D-CNN model with RGB and HS layers produces the highest classification accuracy. The producer accuracy of the best 3D-CNN classifier on the test dataset were 99.6%, 94.8%, and 97.4% for pines, spruces, and birches, respectively.

The best 3D-CNN classifier produced ~5% better classification accuracy than the MLP with all layers. The results suggest that the proposed method provides excellent classification results with acceptable performance metrics for HS datasets. The results show that pine class was detectable in most layers. Spruce was most detectable in RGB data, while birch was most detectable in the HS layers. Furthermore, the RGB datasets provide acceptable results for many low-accuracy applications.