Deep Learning Object Detection Approaches to Signal Identification

TL;DR

This paper reformulates signal source identification as an object detection task using deep learning, demonstrating high accuracy with models like RetinaNet and YOLOv5 trained on a new spectrogram dataset.

Contribution

It introduces a novel approach to source identification via deep learning object detection, along with a new dataset and evaluation of state-of-the-art models.

Findings

Models achieve up to 0.906 Mean Average Precision.

Deep learning models outperform traditional threshold-based algorithms.

The approach effectively detects small and overlapping signals.

Abstract

Traditionally source identification is solved using threshold based energy detection algorithms. These algorithms frequently sum up the activity in regions, and consider regions above a specific activity threshold to be sources. While these algorithms work for the majority of cases, they often fail to detect signals that occupy small frequency bands, fail to distinguish sources with overlapping frequency bands, and cannot detect any signals under a specified signal to noise ratio. Through the conversion of raw signal data to spectrogram, source identification can be framed as an object detection problem. By leveraging modern advancements in deep learning based object detection, we propose a system that manages to alleviate the failure cases encountered when using traditional source identification algorithms. Our contributions include framing source identification as an object detection problem, the publication of a spectrogram object detection dataset, and evaluation of the RetinaNet and YOLOv5 object detection models trained on the dataset. Our final models achieve Mean Average Precisions of up to 0.906. With such a high Mean Average Precision, these models are sufficiently robust for use in real world applications.