top of page

AI Case Study

Wildlife Protection Solutions combats illegal wildlife poaching with deep learning

Wildlife Protection Solutions has leveraged machine learning technology from Silverpond to combat illegal wildlife poaching. The company uses motion detection cameras to detect intruders in conservation parks. The images are then transmitted to the WPS office in real time to be assessed by controllers. With the use of the deep learning system provided by Silverpond, the system, which was trained on classified images of people and objects such as trucks and motorbikes, is now able to detect poaching automatically. The current detection rate is estimated to have increased by 30-40% from only 40% to around 70-80%.


Public And Social Sector


Project Overview

"WPS partnered with Silverpond to implement a solution to automate the detection of illegal poaching in wildlife conservation parks across three continents. Using Silverpond’s cloud-based machine learning technology ‘Silverbrane’, WPS has been able to successfully automate the detection of illegal poaching activity. This has helped the organisation increase efficiencies in their ongoing quest to end wildlife poaching.

Step 1: Understanding the challenges
The Silverbrane team spent time with WPS to understand their unique challenges and requirements. From the consultation period, it became clear that the previous off- the-shelf solutions did not work due to the highly contextual nature of the images produced by the eld cameras.
The priority for the Silverbrane team was to classify images of people and people-related objects, such as trucks and motorbikes. There was also a secondary requirement for various animals to be classiffied to allow the data to be used for future conservation efforts.

Step 2: Annotating the data
The raw data was then uploaded into Silverbrane and
with the assistance of virtual consultation, workshops and training materials, the WPS team was able to annotate the images, identifying people, vehicles and a range of wildlife.

Step 3: Training the deep learning model
With the data set of images now classiffed by WPS, the Silverbrane team set about training the deep learning model. This resulted in a model that was able to easily recognise the classi ed subjects by quickly analysing each image.

Step 4: Evaluating the model
Once the model was complete, an evaluation process was conducted. The evaluation process allows for the e ectiveness of the model to be tested by running new images through it (inference).
While the teams were very pleased with the results, the evaluation indicated that additional annotations for some image classes were required, along with a new classiffcation for ‘antelope’.
Following these additions, the Silverbrane API was implemented successfully into the WPS system."

Reported Results

"Within the first week, it had detected two groups of poachers and automatically sent alerts to Park Rangers, who were able to apprehend them.

Prior to the implementation of Silverbrane, WPS had only a 40% detection rate. Current detection rates are now estimated by WPS
to be 70-80%."






"Wildlife Protection Solutions (WPS) is an international non-profit organisation that provides technology to protect endangered species and ecosystems.

Based in Denver, Colorado, WPS uses technology to conserve endangered species and ecosystems globally. WPS currently provides a fully integrated system for wildlife conservation parks to detect unwanted intruders on their properties. Motion detection cameras transmit images back to the WPS head office in real time. These images are then assessed by a team member and if evidence of people, vehicles or poaching is detected, the appropriate action
is taken.

The requirement of team members to review each image manually was incredibly labour intensive, prompting WPS to look for a solution to streamline the process.

A challenge the company faced was that "the images from which poachers would be detected were not typical and produced in a variety of different environments. For example, they could be taken during the day or night, and from a variety of angles and resolutions based on their placement in the field. These variations meant that off-the-shelf solutions produced inaccurate results."



Images captured by motion detection cameras

bottom of page