A Multimodal UAV-Based Strategy Using Fusion Architecture for Tiger Species Identification in Dense Forest Environments

Abstract

Monitoring wildlife species is important for better ecosystem management and biodiversity conservation. Unfortunately, traditional camera-based wildlife monitoring has limitations, including elevated operational costs, high ecological variability, limited coverage, and false alarms. In this regard, the research details a UAV-based multimodal deep learning approach to detect tigers using high-resolution RGB and multispectral photographs. This system comprises a two-branch architecture: EfficientNet-B3 for spatial feature detection in RGB images, and a lightweight CNN for sensing spectral attributes from multispectral images such as NIR and Red Edge. The feature-level fusion technique combines spatial and frequency features and passes them to the YOLO-based model for detection. Then, U-Net segmentation is used for pixelwise segmentation of pictures. Furthermore, tiger identification is performed using a SoftMax classifier to determine tiger presence and measure confidence. So, the method proposed in this paper is an efficient means for implementing drone monitoring of wildlife species in the real world.