Evaluation of Firefighter Training Effectiveness Based on Human Physiological Signals and Improved Transfer Learning

Worldwide, governments at all levels are trying to minimize the number of firefighter injuries and fatalities during rescue operations. Inadequate day-to-day training has been identified as a significant cause of accidents. Traditional machine learning-based methods to evaluate the training effectiveness of firefighters require large amounts of data. Still, it is difficult to obtain large quantities of data due to the specificity of the firefighting profession and the poor reproducibility of human physiological signals. This study aims to use transfer learning to solve the problem of insufficient sample size resulting in low assessment accuracy. In this study, four human physiological signals surface Electromyography(sEMG), Electrocardiogram(ECG), Photoplethysmography(PPG), and Respiration(RESP) were selected to build a training effectiveness assessment database, using firefighter training data as the target domain and student-simulated firefighter training data as the source domain and a training effectiveness assessment model based on the Improved Joint Distribution Adaptation (JDA) was proposed. Its validity was verified using the public dataset and the self-constructed database. The results show that the accuracy of the improved JDA training effectiveness evaluation model under minor sample conditions is 0.83, which can quickly find the optimal parameters of the model and has higher evaluation accuracy compared with the traditional JDA.