Hybrid Deep Learning Framework for Automated Classification of Wildlife Camera Trap Images

Camera traps play a crucial role in wildlife research and conservation, as they provide invaluable insights into animal behavior and population dynamics. However, the manual analysis of the extensive image data generated by these traps presents a significant challenge. Deep-learning models offer a promising solution for automating image classification and facilitating efficient wildlife monitoring. In this paper, a novel hybrid object classification model for wildlife image classification in camera trap data has been proposed. This model combines the strengths of two well-established deep learning architectures, ResNet-50 and DenseNet-121, with an ensemble-learning approach using logistic regression. The data was acquired from camera traps deployed in wildlife habitats and pre-processed it to enhance image quality and facilitate model training. The pretrained ResNet-50 and DenseNet-121 models on the wildlife image dataset were fine-tuned and the ensemble approach combined the feature vectors extracted by both models and uses logistic regression for final classification. The performance of the model was evaluated using metrics such as $\log$ loss and accuracy. The results show that the proposed hybrid object classification model achieves high accuracy of $\mathbf{89\%}$ in classifying wildife species from camera trap images. Th advances automated wildlife image classification and supports conservation efforts by enabling efficient species identification and population monitoring