*3.4. Experiment of the Entire Network*

Benefiting from the multi-task learning mechanism, the entire network for EV fire trace recognition combined two branches and achieved better performance than using the single severity segmentation branch only. To demonstrate this improvement, different configurations of training and output were implemented using the proposed network, and the results are shown in Table 8 and Figure 10. In the joint training method, the *λ* value for loss calculation was set to 0.25 based on the ratio of the loss value while each branc converged.

**Table 8.** Results of different training methods and output methods. "Branch#2" stands for training the severity segmentation branch only.


**Figure 10.** Results of the proposed DA-EMA in different training and output configurations. (**a**) Original images. (**b**) Labeled images. (**c**) Results of the single severity segmentation branch. (**d**) Results of the two-stage training and 4-class output. (**e**) Results of the joint training and 4-class output. (**f**) Results of the two-stage training and 3-class output. (**g**) Results of the joint training and 3-class output.
