| Forest fires are sudden and strong,and the damage is huge.After the fire occurs,the surface of the forest and the dead wood will remain residual fire.If the residual fire is not removed in time,it is easy to form secondary combustion and further expand the fire range.According to statistics,the occurrence of most forest fires is caused by the lack of timely or incomplete removal of the fire.The fire detection and cleaning work of forest fires is large and difficult,and it has certain threats to the personal safety of firefighters.At present,there are still few studies on the detection and cleanup of forest fires at home and abroad.Based on the current situation of forest fire residual fire detection and cleanup,the paper studies the path planning of forest fire afterfire detection and cleaning robot.On the basis of studying intensive learning and deep learning,the thesis carried out related research work on robot path planning.The main work is as follows.(1)In the study of the characteristics of the reinforcement learning algorithm,theoretically analyze the advantages and disadvantages of different types of reinforcement learning algorithms,based on the principle of Q-learning algorithm to write the program,realize the path planning of Q-learning under simple map.(2)In order to improve the Q-learning algorithm,the “dimension disaster” on the complex map can not effectively carry out path planning,and the deep learning framework is integrated into Q-learning.At the same time,the memory playback matrix and the double-layer network structure are built to improve the convergence of the improved algorithm.(3)Based on the image of mountain road environment,the path planning problem is turned into image recognition problem,and the convolutional neural network is constructed to classify the training samples to realize visual path planning. |
