In order to improve the strike effectiveness of missiles in multi-constraint environments, this paper proposes a guidance law based on improved proportional navigation and deep learning. Firstly, generate the training data and use it to establish a mapping network to predict the error in remaining flight time. Based on this, design a guidance law using an improved proportional navigation method, while both achieving control of the attack angle and attack time under the constraint of the field of view. The numerical simulations demonstrate the effectiveness of the guidance law and show that it has better performance than other methods.