Electric Vehicle Fire Trace Recognition Based on Multi-Task Semantic Segmentation

Conflagration is the major safety issue of electric vehicles (EVs). Due to their well-kept appearance and structure, which demonstrate salient visual changes after combustion, EV bodies are recognized as an important basis for on-spot inspection burnt EVs make application using semantic segmentation possible. The combination deep learning-based recognition traces would provide preliminary analytical results fire spread trends output status descriptions further investigation. In this paper, a dataset image was built, two-branch network structure that splits whole task into two sub-tasks separately concentrated on foreground extraction severity proposed. proposed trained via transfer learning method tested 5-fold cross validation. branch achieved mean intersection over union (mIoU) 95.16% in task, mIoU 66.96% task. By jointly training branches applying mask 3-class output, improved 68.92%.