Abstract: To address the difficulty of non-contact estimation of internal heat release rate (HRR) in historical building fires,a method for predicting transient HRR from external temperature field images was proposed.The fire dynamics simulator (FDS) was used to construct 60 fire scenarios for historical buildings with 45° pitched roofs under different fire source intensities and soot yield values.External temperature field images were extracted from front,side,and top views and matched with the corresponding transient HRR values to establish a regression dataset linking multi-view images and HRR.The TripleCNN-SE model was trained on this dataset for multi-view feature fusion and transient HRR prediction.The results show that the model achieved a validation coefficient of determination of 0.982 and a mean relative error of 7.8%.For fast-growing fire scenarios and scenarios with fuel parameters excluded from training,the mean relative error remained below 20%.These results indicate that,under controlled simulation conditions,indoor fire HRR in historical buildings can be inferred from external temperature field images,providing a methodological basis for non-contact fire condition sensing and early warning research in historical building fires.