A Wildlife Multi-Modal Recognition Method Based on Infrared Camera Images Using MWAC

Nowadays, infrared camera technology has become an important tool for large-scale monitoring and assessment of wildlife and has generated a huge amount of image data, but how to quickly identify wildlife species has become a major problem nowadays. To solve this problem, this study proposes MWAC, a model for accurate classification of wildlife species in infrared camera images by integrating image and text descriptions. First, we propose a wavelet KAN module (WDS-KAN) for multi-scale feature extraction of wildlife images, which enables the model to capture the subtle changes in animal images more accurately. Second, a label-correlated text extraction module (LC-BERT) is proposed to capture label co-occurrence patterns and hierarchical relationships in the dataset, which enhances the generalization ability of the model. Finally, a multi-strategy coati optimization algorithm (MSCOA) is proposed, which enhances our ability to globally search for the optimal solution, effectively avoids falling into local optima, and accelerates the model training speed. In order to verify the effectiveness of the model, we constructed a dataset containing 10 species of wild animals. The experimental results show that the MWAC model achieves 97.53% in accuracy, 97.33% in precision, 95.84% in recall, and 97.51% in F1, which are better than those of the existing models, indicating that the model is good for the rapid identification of wildlife and other aspects of good results and can be applied to large-scale wildlife monitoring and assessment.