Real-Time Localization and Bimodal Point Pattern Analysis of Palms Using UAV Imagery

Kangning Cui, Wei Tang, Rongkun Zhu, Manqi Wang, Gregory D. Larsen, Victor Paúl Pauca, Sarra M. Alqahtani, Fan Yang, David Segurado, Paul Fine, Jordan Karubian, Raymond H. Chan, Robert J. Plemmons, Jean-Michel Morel, Miles R. SilmanarXiv 2024 · DOI · arXiv

Abstract

Accurate monitoring of palm populations in natural tropical forests is critical for conservation but hindered by complex canopy structures and computational constraints. This paper presents PalmDSNet, a deep learning framework designed for real-time detection, segmentation, and counting of palms using Unmanned Aerial Vehicle (UAV) imagery. By integrating state-of-the-art object detection models with a novel Poisson-Gaussian bimodal reproduction algorithm, the study enables both precise mapping and ecological simulation of palm spatial distributions.

Introduction

Monitoring tropical forests is notoriously difficult. Unlike structured palm oil plantations, natural forests are characterized by overlapping crowns, heterogeneous terrain, and complex lighting conditions. These factors create "false" features that challenge traditional remote sensing methods. Furthermore, the need for real-time processing on-board UAVs necessitates lightweight yet accurate models that can operate in chaotic environments.

Detection of overlapping palm crowns in dense tropical forest canopies, illustrating the complexity of natural forest monitoring.

This research addresses these challenges by providing a robust dataset of over 7,000 expert-annotated palms across diverse ecological gradients in Ecuador and a flexible framework that bridges high-accuracy deep learning with practical field ecology.

Methodology

The proposed solution consists of two primary components:

  1. PalmDSNet Framework: A modular architecture that utilizes detection backbones—including YOLOv8, v9, v10, and RT-DETR—to localize palms. This is followed by the Segment Anything Model (SAM) for precise crown delineation. To ensure model interpretability, Grad-CAM saliency maps were employed to visualize the features (e.g., leaf patterns) driving the AI's predictions.
  2. Bimodal Reproduction Algorithm: To move beyond simple counting, the authors developed a simulation model to analyze spatial distribution. This algorithm combines a Poisson process (simulating long-range animal-mediated seed dispersal) with a local Gaussian distribution (simulating seeds falling near parent trees). The simulations were validated against real-world data using Ripley’s K-series functions.

Data and Experiments

The study utilized UAV-captured imagery from 21 sites in western Ecuador, representing a gradient from everwet rainforests to dry tropical forests. The resulting dataset includes 7,356 bounding boxes and 7,603 manually marked palm centers, providing a new benchmark for the community.

Palm detection performance across diverse ecological gradients in western Ecuador, showing the model's robustness in various forest types and densities.

Key Results and Insights

  • Model Performance: RT-DETR and YOLOv9 emerged as the strongest models. RT-DETR achieved a Mean Average Precision (mAP) of 0.734 and an AP75 (a stricter accuracy metric) of 0.822 when trained on 90% of the dataset.
  • Counting Accuracy: The framework demonstrated exceptional reliability, with counting accuracy exceeding 99% in environments similar to the training data and remaining above 94% in previously unseen forest types.
  • Real-Time Efficiency: Inference speeds exceeded 40 frames per second (FPS) on standard hardware, confirming the feasibility of on-board UAV deployment for rapid ecological assessments.
  • Ecological Findings: Spatial analysis confirmed that palms are not randomly distributed but exhibit clustering and "negative density dependence." The success of the bimodal model suggests that palm populations are shaped by a balance of local recruitment and long-distance dispersal.

Conclusion and Future Work

PalmDSNet provides a scalable, real-time solution for forest monitoring. While currently robust, future work will focus on localized parameter estimation for the bimodal algorithm and the use of raw multi-view UAV imagery to detect palms partially obscured by taller canopy layers. Expanding the dataset to include species-level identification remains a priority for tracking endangered or economically vital species across the Amazonian basin.

Summary updated

Abstract

Understanding the spatial distribution of palms within tropical forests is essential for effective ecological monitoring, conservation strategies, and the sustainable integration of natural forest products into local and global supply chains. However, the analysis of remotely sensed data in these environments faces significant challenges, such as overlapping palm and tree crowns, uneven shading across the canopy surface, and the heterogeneous nature of the forest landscapes, which often affect the performance of palm detection and segmentation algorithms. To overcome these issues, we introduce PalmDSNet, a deep learning framework for real-time detection, segmentation, and counting of canopy palms. Additionally, we employ a bimodal reproduction algorithm that simulates palm spatial propagation to further enhance the understanding of these point patterns using PalmDSNet's results. We used UAV-captured imagery to create orthomosaics from 21 sites across western Ecuadorian tropical forests, covering a gradient from the everwet Chocó forests near Colombia to the drier forests of southwestern Ecuador. Expert annotations were used to create a comprehensive dataset, including 7,356 bounding boxes on image patches and 7,603 palm centers across five orthomosaics, encompassing a total area of 449 hectares. By combining PalmDSNet with the bimodal reproduction algorithm, which optimizes parameters for both local and global spatial variability, we effectively simulate the spatial distribution of palms in diverse and dense tropical environments, validating its utility for advanced applications in tropical forest monitoring and remote sensing analysis.
BibTeX
@misc{cui2024real,
  author = {Kangning Cui and Wei Tang and Rongkun Zhu and Manqi Wang and Gregory D. Larsen and Victor Paúl Pauca and Sarra M. Alqahtani and Fan Yang and David Segurado and Paul Fine and Jordan Karubian and Raymond H. Chan and Robert J. Plemmons and Jean-Michel Morel and Miles R. Silman},
  title = {{Real-Time Localization and Bimodal Point Pattern Analysis of Palms Using UAV Imagery}},
  year = {2024},
  eprint = {2410.11124},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2410.11124},
}
Figure