Abstract Ultra‐high‐voltage direct current wall bushings are critical components in direct current transmission systems. Temperature variations and abnormal distributions can signal potential equipment failures that threaten system stability. Therefore, monitoring these critical multi‐point temperature variations is essential. However, the unique design of the bushings, featuring insulation sheds of periodic shape, distorts infrared temperature measurements by introducing interference points. These interference points, dependent on the measurement's angle and distance, appear irregularly in infrared images, severely impacting the accuracy of multi‐point temperature distribution assessments. To address this challenge, an anomaly detection method is proposed that adaptively identifies interference points. The method identifies interference points by comparing pixels and uses a voting mechanism to improve identification accuracy. Compared with traditional methods, this approach presents two main advantages: adaptive identification capability, which enables it to recognise interference points and adapt to changing conditions, and unsupervised learning, which enables it to work effectively without requiring manually labelled data. Experimental tests on 161 bushing infrared images demonstrate the effectiveness of the method, achieving a 100% success rate in identifying localised overheating issues. The method has been integrated into high‐voltage direct current transmission anomaly systems and can be used to monitor critical equipment, enhancing system reliability and safety.