3S Forest Fire Prevention Construction Intelligent System based on Communication System Optimization Model

Forest is the main body of China's ecological construction and an indispensable part in the realization of China's sustainable development strategy. As of September 2017, China's forest area has reached 205.8 billion hectares, and the forest coverage rate has reached 21.93%. At the same time, as a big country of forest resources, forest fires occur frequently. This paper proposes hourglass architecture and sparse gradient compression from two aspects of system architecture and communication data compression to reduce the communication overhead of distributed training deep neural network. The algorithm is used to speed up the training process to ensure the accuracy loss within 1%.