WILD ANIMAL ACTIVITY ALERTS WITH HYBRID DEEP NEURAL NETWORKS

Abstract

People who work in forestry and reside in rural areas are growing increasingly concerned about animal attacks. Drones and surveillance cameras are frequently employed to monitor and track wild animals. However, you need a competent model in order to identify the type of animal, monitor its activity, and determine its position. Then, in order to protect people and foresters, alert signals might be sent out. Animals are frequently located using computer vision and machine learning-based techniques, but these approaches are typically expensive and difficult to deploy, which makes it difficult to achieve good results. This study presents a Hybrid Visual Geometry Group (VGG)−19+ Bidirectional Long Short-Term Memory (Bi-LSTM) network for animal detection and activity-based alert generating. To ensure prompt action, these notices are sent to the local forest office via Short Message Service (SMS). With an average classification accuracy of 98%, a mean Average Precision (mAP) of 77.2%, and a Frame Per Second (FPS) of 170, the proposed model demonstrates significant performance gains. Using 40,000 images from three different benchmark datasets with 25 classes, we assessed the model. The model's average precision and accuracy were higher than 98%. This model is a reliable method for obtaining accurate animal information and ensuring public safety.