Infrared Heating Anomaly Detection Method Based on Multi-Modal Data Fusion

This paper aims to provide a defect detection method based on multi-modal image fusion and lightweight target detection. The method first collects images by a drone equipped with an infrared camera and an RGB camera. Subsequently, the defect area in the image is marked by manual labeling tools to construct a defect data set. Combined with prior knowledge, a target detection model suitable for defect detection is designed. The model optimizes the backbone network and the detection head part of the model by improving the target detection model YOLOv11. PKINet is used as the backbone and LSCD is used as the detection head to improve the detection robustness and accuracy. Then, the model is used to detect the infrared abnormal heating defect area in the image, and the infrared abnormal heating is extracted by MCMLSD method for further cross-validation and defect location. By constructing an infrared heating anomaly data set for experiments, mAP is increased by 1.5 % on the experimental data set, which verifies that this method can improve the detection accuracy and improve the efficiency and reliability of overhead line defect detection under the premise of ensuring low computing resource consumption.