Exo-atmospheric infrared objects classification using recurrence-plots-based convolutional neural networks.

Area of Science:

• Aerospace Engineering
• Computer Vision
• Signal Processing

Background:

• Object discrimination is crucial for infrared (IR) imaging systems, but classifying exo-atmospheric objects at long distances is challenging due to detector noise and limited features.
• Existing methods struggle with the complexities of IR signatures from objects in space, especially concerning variable data lengths and subtle dynamic changes.

Purpose of the Study:

• To propose a novel Recurrence Plots-based Convolutional Neural Network (RP-CNN) for robust feature learning and classification of exo-atmospheric objects using IR radiation data.
• To leverage the unique textural and graphical properties of Recurrence Plots (RPs) for enhanced analysis of time-evolving IR signatures.
• To improve the accuracy and robustness of classifying objects in space from long-range IR observations.

Main Methods:

• Transforming time sequences of IR radiation into 2D texture images using Recurrence Plots (RPs).
• Employing a Convolutional Neural Network (CNN) model for classification based on the generated RP images.
• Generating training data using IR irradiation models that incorporate micro-motion dynamics and geometrical shapes of exo-atmospheric objects.

Main Results:

• The proposed RP-CNN method demonstrates significant improvements in classification accuracy compared to existing approaches.
• The technique shows enhanced robustness in classifying exo-atmospheric objects, even with noisy IR data and varying object signature lengths.
• RP representation effectively reveals hidden patterns and structural changes within IR signatures, aiding in better discrimination.

Conclusions:

• RP-CNN offers a powerful and effective approach for the classification of exo-atmospheric objects based on IR signatures.
• The method's ability to handle variable-length data and its robustness to noise make it suitable for long-distance space surveillance.
• This research advances the field of space object identification by providing a more accurate and reliable classification technique.