Immersive experiences in film and TV depend largely on spatial audio handling. The existing methods lack accuracy and efficiency in handling complex sound fields. In this paper, deep learning-based spatial audio handling for TV and film is introduced, where an audio 3D localization network and an analysis framework for training the sound field reconstruction model were implemented. It employs a standardized process in collecting data and extracting multidimensions for modeling parameters in the spatial manner. Experiments were performed using several environmental and multi-source audio samples, with error analysis and subjective scores being used in assessing the effect on sound field reconstruction and area localization. The findings indicate superior performance compared to conventional methods in the accuracy in placing locations and quality in perception in the spatial manner, as evidenced by having less than 4 degrees in the mean angle error and high increases in the sound field immersion test scores. This proves the viability and application value of the method in promoting the quality of the spatial handling effect.