Deep Learning-Based Object Detection in Thermal Imaging

Thermal imaging is crucial for environmental monitoring, industrial inspection, and security, yet object detection in thermal images faces challenges like low contrast, noise, and temperature variations. This paper presents a deep learning-based approach integrating YOLOv5 and Faster R-CNN for object detection, alongside autoencoder-based anomaly detection for identifying temperature deviations. The model is trained on FLIR Thermal and KAIST Multispectral datasets, achieving a mAP of 0.973 at an IoU threshold of 0.5, with precision (0.949) and recall (0.921) outperforming conventional methods. The framework supports real-time processing, enhancing industrial safety and environmental monitoring. Future research will explore multi-modal data fusion, Vision Transformers (ViTs), and edge computing for scalable deployment. This study emphasizes the potential of deep learning to transform thermal imaging applications.