基于多传感器数据融合的早期林火识别

针对我国林火监测现状，为加强近地面监测中的早期林火发现，提出采用多传感器数据融合算法对早期林火进行识别的方法。通过设计林火仿真试验，采集CO₂浓度、CO浓度、烟雾浓度与空气温湿度等多传感器数据，并通过初步分析从中选取关键贡献率传感器数据。然后分别采用BP神经网络算法、神经模糊系统算法与支持向量机算法对数据进行识别与分析, 并在每个算法中均设置三输入与九输入2种不同输入向量数以进行比较。最后通过定义的识别性能评价参数对识别效果进行比较，得出支持向量机算法在一定范围内能较好地实现对早期林火的识别。

Early Forest Fire Detection Algorithm Based on Multi-sensor Data Fusion

In order to improve the detection of early forest fire monitor near the ground, this paper proposed three algorithms of multi-sensor data fusion after analyzing the domestic status of forest fire monitoring. In the simulation experiment of forest fire, the data which included concentrations of carbon dioxide, carbon monoxide, smoke, air temperature and humidity were collected. After being analyzed preliminarily, the data of the mainly contributive sensors were picked out. And then three algorithms, which included BP neural network, fuzzy neural system and support vector machine were used to analyze the experiment data and identify the fire status. In the algorithms, three input vectors and nine input vectors were both set as inputs for comparison. At last three parameters of identification performance were defined to compare the identification’s performance of the three algorithms. The conclusion was that the algorithm of support vector machines was able to perform a good identification of the early forest fire within a certain range.