Plume Source Target Extraction And Autonomous Identification Methods

With the progress of science technology and industry,many dangerous chemicals are more and more widely used in human production and life.When natural and manmade disasters occur,many people will be participate in on-site rescue.Especially in the production and manufacturing process of chemical plants,the plume source leakage occurs in many cases all over the world.This will undoubtedly cause tremendous loss of human and financial resources.Therefore,when the plume source leaks,it is important to find the plume and locate the plume in time to reduce the loss.The main contents of this paper are target extraction and self-identification of plumes.On the basis of the previous research results,the plume identification is further studied.Traditional plume tracing only uses olfactory sensor and wind direction sensor for plume source searching.This paper proposes a fusion of visual and olfactory sensors for plume source searching,which reduces the type and number of sensors,and achieves a better effect of plume source identification.The database of plume and interferences is constructed.After normalization,the image data are preprocessed to extract gray and shape features of plume source.The extracted feature images are classified by putting KNN algorithm,BP neural network and SVM.The predicted labels are compared with the known image class labels,and the correct rate of the test set is obtained.However,in plume feature extraction and target recognition,the accuracy of test set is not high and plume classification is difficult to accurately classify.In view of the above problems,the convolution neural network is used in this paper,and the proposed improved convolution neural network is used to extract and identify plume targets.In the improved convolutional neural network,combining migration learning method,and replacing the original Softmax classifier with SVM classifier to improve the accuracy of experimental test set.The experimental results show that the improved convolution neural network has higher recognition accuracy for different types of plumes,and the number of iterations is reduced.Therefore,it can achieve higher recognition rate in shorter training time and has good robustness.Through the establishment of Simulink simulation platform,the plume source search process is simulated,and the motion process parameters are analyzed,and good application results are obtained.Finally,the experimental platform is built to further verify the effect.The experimental platform uses TurtleBot mobile robot and the plume sensor module to collect plume information,which is verified by the above methods.In this experiment,visual information can not be used to determine which plume is leaking.This paper collects plume information by fusing olfactory sensors,which plays an auxiliary role in visual plume sourcing and achieves good application results.