In a thermoforming process in which a steel plate is formed into a curved shape with line heating, it is not easy to select heating positions to obtain the desired shape, so it mainly depends on the empirical knowledge of skilled workers. In this study, using Faster R-CNN, a deep learning model was proposed that takes the desired surface as an input and the heating positions to obtain it as an output. The data sets for training the model were obtained by using a finite element analysis model that can predict the deformation according to the heating process parameters. The model was trained by setting color map images of various deformed shapes obtained from the analysis as input data sets and heating positions that caused deformations as output data sets. Using the trained model, an arbitrary deformed shape image was input, and as a virtual object existing in the image, the positions of the heating lines could be predicted. As a result of the test, it was shown that the proposed model predicts the heating lines very accurately. In a validation of the proposed model, the performance of accurately predicting the heating positions for forming the desired shape was shown.