Measuring and Predicting Sensor Performance for Camouflage Detection in Multispectral Imagery.

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Effective management of multispectral sensor systems on small reconnaissance drones is crucial for target detection.
• Identifying camouflaged targets requires optimizing sensor band selection based on environmental context.

Purpose of the Study:

• To propose an approach for predicting multispectral sensor band performance in exposing camouflaged targets.
• To introduce a new metric, the Target Visibility Index (TVI), for quantifying target visibility.
• To rank sensor bands based on predicted performance for improved reconnaissance.

Main Methods:

• Developed a new metric, Target Visibility Index (TVI), to quantify camouflaged target visibility in specific sensor bands.
• Trained machine learning models to predict TVI based on environmental context extracted from visual bands using image descriptors.
• Created and utilized a dataset of 853 multispectral captures with diverse camouflaged targets and environments for model training and evaluation.

Main Results:

• The proposed approach successfully predicted the most informative sensor bands in most tested scenarios.
• Machine learning models effectively learned the relationship between TVI and environmental context.
• The developed dataset is publicly available, facilitating further research.

Conclusions:

• The performance prediction approach significantly enhances camouflage detection in real-world reconnaissance.
• It improves the utility of individual sensor bands and reduces workload for complex multispectral systems.
• This method offers a valuable tool for optimizing drone-based surveillance and target identification.