# Detect and Trace: An Australian Field Trial Using Machine-Learning Tools to Combat Illegal Wildlife Trade

## TL;DR

This paper describes a trial in Australia using machine learning to help law enforcement intercept illegal wildlife trade, showing promising results in detecting smuggled animals.

## Contribution

The study demonstrates the practical application of machine learning tools in real-world wildlife seizures and their impact on reducing illegal trade.

## Key findings

- Machine learning detected smuggled wildlife in 56% of scanned parcels using the AT.3 algorithm.
- X-ray imaging captured 48 high-resolution 3D images for identifying concealed wildlife.
- Provenance models helped differentiate between wild-caught and captive-bred lizards.

## Abstract

While efforts are underway globally to combat illegal wildlife trade through a coordinated approach, testing of new technologies in real-world settings remains limited. Here, we present the outcomes of an opportunistic Australian trial that tested two machine-learning tools during real-world seizures, including associated radiation-exposure safety data. We demonstrate that machine learning technology can have a meaningful impact when used alongside law enforcement to reduce the illegal trade of wildlife.

The illegal wildlife trade is a global problem that continues to harm individuals, wildlife populations, ecosystems, and humans at an increasing rate. While efforts are underway globally to address the issue through a coordinated approach, the testing of new technologies in real-world settings remains limited. Here, we present the outcomes of an opportunistic Australian trial that tested two machine-learning tools during real-world seizures, including associated radiation-exposure safety data. During the seven-month trial, 116 animals were intercepted, representing reptiles and crustacea across five Genera: Tiliqua, Egernia, Oedura, Chelodina, and Euastacus. Of the 18 seized consignments, totalling 48 parcels, scanned through the RTT®110 CT X-ray baggage scanner, automated AI detected smuggled wildlife 56% of the time using the most successful algorithm (AT.3), and captured 48 high-resolution 3D X-ray images, which allowed identification of concealed wildlife. In addition, 33 Tiliqua sp. were scanned using the Olympus Vanta pXRF and the data analysed using previously published machine-learning provenance models. Common blue-tongue lizards (Tiliqua scincoides) were less likely to be wild-caught than shingleback lizards (Tiliqua rugosa). Alongside expert statements, provenance results were provided to enforcement agencies. Following the trial, there was a significant reduction in the number of seized parcels being exported through postal pathways. This trial demonstrates the impact of integrating new technology to support intelligence-led enforcement processes and reduce wildlife trafficking.