Developing hybrid deep neural networks for detecting the movement of wild animals and generating alarm messages

Abstract

People living in rural areas and those who work in the forest are becoming more and more worried about animal assaults. It is common practice to use cameras for surveillance and drones to monitor the whereabouts of wild animals. The animal's kind, its motility, and its position may all be detected with the use of an efficient model. In order to guarantee the safety of both people and foresters, alert messages might subsequently be delivered. Although methods based on image recognition and machine learning are often used for animal identification, they may be rather costly and intricate, which can make it challenging to get satisfying outcomes. Animals may be identified and alarms can be generated according to their actions using a network called Bi-LSTM, which is a Hybrid Visual Geometric Group (VGG) −19+ network. The local forestry office receives these notifications via SMS, or short message service, so they can respond immediately. With an overall Average Accuracy (map) of 77.2%, a frame rate Per Seconds (FPS) of 170, and an average accuracy of Classification of 98%, the suggested model shows significant increases in model performance. Using 40,000 photos across three distinct data sets with 25 classes, the model underwent quantitative and qualitative testing, and it attained an average precision and precision of over 98%. If we want trustworthy data derived from animals that won't endanger people, this methodology is the way to go.