DEEP LEARNING OF QINLING FOREST FIRE ANOMALY DETECTION BASED ON GENETIC ALGORITHM OPTIMIZATION

Located in the most famous part of the north-south boundary of China, Qinling Mountains boast dense forests and high vegetation coverage. In forest areas, effective monitoring and prevention of fires is an important issue demanding urgent solution. This is because fire prevention can not only effectively protect the surrounding environment, but also provide an important guarantee for living and personal safety of surrounding residents. To achieve rapid, effective and accurate analysis and prevention of forest fires, this paper proposes a deep learning timeseries convolutional neural network (GA-CNN) optimized by genetic algorithm, which can detect fire occurrences with high accuracy in the environment. In different evaluation scenarios, it exhibits good performance in accuracy, true positivity, and false alarm rate. The optimized CNN is not only capable of global optimization, but also can judge convolutional neural network by time series. In this way, it can not only avoid the problem of convergence difficulty and improper model structure selection that are often encountered in convolutional neural networks, but also effectively reduce the allocation of human resources for forest protection. Experimental results show that the proposed forest fire image recognition method achieves higher recognition and false alarm rates than other algorithms (BP (Back Propagation) neural networks, namely the learning process of error inversion error backpropagation algorithm, SVM (Support Vector Machine) and genetic algorithm optimized BP neural network (GA-BP) and can be used to liberate manpower without requiring complex manual feature extraction to identify features of forest fire edges.