A Machine Learning Framework for Combatting Poaching Challenges in Wildlife

Global threats to wildlife pose a major challenge, as the natural habitats of animals in Africa and other regions are frequently targeted by poachers and hunters with ulterior motives, resulting in the extinction of certain species. This research presents a combined machine learning framework for detecting poaching through forest video analysis. The framework is based on three models: Deep Residual Convolutional Neural Networks (DRCNN) for detecting and identifying wildlife, a Residual Network (ResNet) for recognizing their action and/or movements, and a customize CNN combined with a Support Vector Machine (SVM) for sound classification. This system seeks to promote nature conservations, by identifying animals, their actions, and sounds to assess the likelihood of threats to wildlife. The proposed model possesses enhanced abilities to detect animal actions and sounds under stress, compared to conventional wildlife guard detection methods, which often overlook the impact of animal behaviour interference in wildlife parks. The proposed model further considers information technology tools and video recorded from Unmanned Aerial Vehicles (UAVs) or other sources to provide a more precise assessment of wildlife, emphasizing the greater intervention required to enhance monitoring of wildlife conservation areas. In order to implement this framework, a standard dataset is essential for specific wildlife actions and their usual sounds they make based on the context or the activities the animal is engaged in. The framework is universal to all animals, based on their typical behaviors/actions and vocalizations according to the context.