Integrating Deep Learning and Spatial Statistics in Marine Ecosystem Monitoring

2025-11-20

AI Summary

Automatic detection of benthic marine organisms (e.g., sea cucumbers) suffers from spatially varying detection probabilities, leading to biased estimates of their true spatial distributions. Method: We propose a correction framework integrating deep learning with spatial statistics, built upon a thinned Log-Gaussian Cox Process (LGCP) model. This model treats deep learning detection outputs as biased, spatially heterogeneous observations and explicitly incorporates detection probability heterogeneity, calibrated using sparse human annotations. Results: Evaluated on an underwater image dataset from the coastal waters near Giglio Island, Italy, our method significantly improves reconstruction accuracy of true species distributions and reduces density estimation error compared to conventional LGCP and raw detection outputs. This work represents the first systematic application of thinned LGCPs in marine photogrammetric ecological monitoring, establishing a scalable, automated, and statistically robust modeling paradigm for large-scale benthic biodiversity surveys.

Technology Category

Machine Learning: Calibration & Uncertainty QuantificationReasoning under Uncertainty: Relational Probabilistic ModelsComputer Vision: Biometrics, Face, Gesture & Pose

Application Category

Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsWeb Mining and Content Analysis: Large pretrained models with web dataEconomics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

In ecology, photogrammetry is a crucial method for efficiently collecting non-destructive samples of natural environments. When estimating the spatial distribution of animals, detecting objects in large-scale images becomes crucial. Object detection models enable large-scale analysis but introduce uncertainty because detection probability depends on various factors. To address detection bias, we model the distribution of a species of benthic animals (holothurians) in an area of the Italian Tyrrhenian coast near Giglio Island using a Thinned Log-Gaussian Cox Process (LGCP). We assume that a "true" intensity function accurately describes the distribution, while the observed process, resulting from independent thinning, is represented by a degraded intensity. The detection function controls the thinning mechanism, influenced by the object's location and other detection-related features. We use manual identification of holothurians as our benchmark. We compare automatic detection with this benchmark, an unthinned LGCP, and the thinned model to highlight the improvements gained from the proposed approach.Our method allows researchers to use photogrammetry, automatically identify objects of interest, and correct biases and approximations caused by the observation process.

Problem

• Modeling spatial distribution of marine species using thinned point processes
• Correcting detection bias in automated object identification from images
• Integrating deep learning with spatial statistics for ecological monitoring

Innovation

• Integrates deep learning with spatial statistics modeling
• Uses Thinned Log-Gaussian Cox Process for bias correction
• Combines automatic detection with manual benchmark validation