Multispectral representation of Distributed Acoustic Sensing data: A framework for physically interpretable feature extraction and visualisation

Distributed Acoustic Sensing (DAS) enables continuous monitoring of dynamic strain along tens of kilometres of optical fibre, generating massive datasets whose interpretation and automated analysis remain challenging. DAS data can be understood as a time-distance image array. However, the absence of standardised representation methods limits both the physical interpretability and the development of automated image-based analysis pipelines. This work introduces a systematic framework for visualisation and feature extraction of DAS data based on a multispectral signal representation. The approach decomposes strain-rate measurements into predefined frequency bands and produces band-limited energy images that describe the spatial and temporal distribution of acoustic energy across distinct spectral regimes. The framework is evaluated using DAS recordings containing fin whale and blue whale vocalisations. Three experiments are conducted to assess the approach: enhanced visualisation of bioacoustic signals, unsupervised clustering of acoustic patterns, and event detection using a convolutional neural network. The visualisation experiment shows that the multispectral representation allows for the simultaneous visualisation of sets of events that cannot be visualised using the strain amplitude representation. The unsupervised clustering experiment confirms that the multispectral representation yields class-discriminative feature spaces without requiring labelled data, enabling the automatic grouping of acoustically distinct event types. Finally, the use of multispectral composites as input of a ResNet-18 achieves an accuracy of 95.9% in whale vocalisation detection, demonstrating that the proposed representation provides an effective feature space for automated analysis. This approach provides a generalisable bridge between raw sensor data and standard computer vision pipelines, with implications for large-scale automated monitoring using fibre-optic sensor networks.