It has become the research focus to realize the infrared radiation spectrum regulation by designing and optimizing micro/nano photonic structures. Recently, neural network inverse design has attracted wide attention because of its advantages such as high freedom, fast speed and good performance. Here, infrared radiation computation and inverse design of micro/nano photonic structures based on neural network is proposed. Specifically, for the multilayer dielectric thin film structure, a multilayer perceptron neural network model was established. The mapping relationship between the thickness of the multilayer thin film and its infrared radiation spectrum was built up through the training of sample data. The radiation spectrum computation and inverse design of the thin film structure by this network was realized. At the same time the designed film was applied to the field of radiative cooling. The results show that even at 3% solar radiation absorption and 6 W/(m²·K) non-radiation heat transfer coefficient, the thin film radiative cooler can still reduce the temperature to about 7 ℃ below the ambient temperature. The research results will have an important impact on infrared radiation and other applications.