Forest fire prevention is the basic guarantee for building ecological civilization. At present, the main problems of forest fire monitoring based on video are that the algorithm recognition generality is not strong, the algorithm has poor real-time performance, and the rate of missing and false alarm is relatively high. In order to improve the fire recognition rate and real-time, in this project we intend to carry out the following research work: (1) using the ratio of infrared radiation energy of adjacent channels to determine the temperature of ignition point hidden under vegetation in forest areas, screening for the image feature processing of smoke flame in the later stage;(2) combining with infrared spectral feature analysis, the image color characteristics, area characteristics and roundness of forest smoke and flame are studied, and a multi-feature analysis algorithm is established;(3) to build a monitoring platform for forest fire detection based on infrared and visible binocular vision. According to the different characteristics of infrared and visible images, the multi-feature recognition algorithm of video images and the infrared spectral features are studied to detect whether the current monitoring area is abnormal in temperature, smoke or flame. Finally, the ground coordinates of current fire alarm are located by GIS map data, so as to realize the early automatic alarm display of fire situation. Therefore, the solution of forest fire compound feature recognition algorithm based on infrared and visible light can realize the early recognition of forest fire, reduce disaster loss, and help to promote the construction of ecological civilization in China.