Classification and identification of major hazard installations in unattended AC/DC joint construction station based on improved YOLOv3 – DN algorithm

To avoid economic losses in high cost unmanned AC/DC joint construction stations, capture detailed information about station status, abnormal environmental changes, and other details, the depth extraction of image features should be realized, and the classification and identification accuracy of major hazard sources should be improved. A method based on the improved YOLOv3 – DN algorithm is researched for this purpose. This method analyzes and classifies major hazard installations, uses high-definition infrared camera to collect images, and inputs them into the improved YOLOv3 – DN algorithm model. The model introduces a channel attention model into the DenseNet 160 structure, extracts the depth characteristics of the infrared image, and then use the YOLOv3 algorithm to reverse the calculation process, better understanding feature information. The classification value of major hazard installations is taken as the identification limit, outputting classification results. Experiments show that this method has a strong ability to obtain the infrared image features of major hazard installations in unattended AC/DC joint construction stations and can effectively obtain the classification results of major hazard installations in unattended AC/DC joint construction stations. Simultaneously, the improved YOLOv3 – DN algorithm model has a low loss value and good application effect.