A novel thermal image based cold object detection and classification using machine learning algorithms.

Cold object detection and classification allow reliable identification and categorization of low-temperature objects using non-contact thermal analysis, thereby complementing standard heat-focused techniques. It supports accurate monitoring of cooling behaviour, improves fault detection, quality control and remains powerful under poor lighting or visually challenging situations. However, cold object classification remains challenging when using conventional visible-light images due to the absence of discriminative temperature information. As of now, dedicated and systematic research on cold object detection and classification using thermal images are still very limited, highlighting a clear research gap. To address this limitation, this work proposes a thermal image-based cold object detection and classification framework using machine learning algorithms, designed to achieve effectiveness, reduced cost, and lower processing time. In addition, a dedicated dataset is developed by capturing time-dependent temperature variations of multiple cold object categories, enabling well ordered analysis and classification based on their thermal behaviour. Several machine learning models, including Decision Tree, Random Forest, XGBoost, and other classification algorithms, were developed and assessed using the proposed dataset. Among these models, the Random Forest classifier gives the highest classification accuracy of 99.35%, demonstrating its effectiveness in capturing temporal thermal variations for accurate cold object detection and identification. This work establishes a new research direction in thermal image analysis by shifting the focus from heat anomaly classification toward reliable cold object classification.