Artificial Perceptual Learning (APL)

Introduction

This repo includes the sample data and codes for the paper Large-scale, image-based tree species mapping in a tropical forest using artificial perceptual learning.

FIGURE 1. Comparison between APL and infant categorization.

The above image illustrates the comparison between artificial perceptual learning (APL) and infant categorization. The proposed APL framework is constructed using state-of-the-art machine learning algorithms as building blocks that mimic the cognitive development process known as infant categorization.

Implementation for case study

We used the APL framework to estimate the spatial distribution of three target species, the palm Prestoea acuminata, and the tree species Cecropia schreberiana and Manilkara bidentata, over a 5,000-hectare area of El Yunque National Forest.

FIGURE 2. Left: Location of El Yunque National Forest (EYNF) in Puerto Rico; Right: G-LiHT flight lines and location of the Luquillo Forest Dynamics Plot (LFDP) (yellow box).

The large-scale maps are based on unlabeled high-resolution aerial images of unsegmented trees. Misaligned ground-based labels, available for less than 1% of these images, serve as the only weak supervision. The whole workflow for the species distribution estimation is illustrated in Figure 3, with more details introduced in the paper.

FIGURE 3. Diagram for APL and basic workflow.

Visualization for species distribution estimation

FIGURE 4. Visualization of APL predictions over sample images. Left: Prestoea acuminata; Center: Cecropia schreberiana; Right: Manilkara bidentata. The color represents the estimated proportion covered by the target species (Blue: low, Red: high).

FIGURE 5. Visualization of APL predictions over EYNF. Upper Left: Prestoea acuminata; Upper Right: Cecropia schreberiana; Lower Left: Manilkara bidentata. The color represents the estimated proportion covered by the target species (Blue: low, Red: high).

Folder Setup

• /data/: sample data used in the analysis.
• /doc/: IPython notebooks to reproduce the results.
• /figs/: figures used in the README.md.
• /libs/: function definitions.
• /output/: results of the analysis, generated by the codes in /doc/.