Clustering Algorithm And Hierarchical Map Creation Based On Scale Division And Region Growth

At present,the research on harmful gas plume detection is relatively few,and there are some defects: assuming that bottles,pipes,etc.are suspected leakage sources(in fact,the shape and category of suspected leakage sources are uncertain);limiting the information such as wind speed and gas distribution;the real-time performance of harmful gas plume detection is poor.In order to improve the universality of the plume detection robot,it is necessary to solve the problem that the shape and type of the suspected leakage source are uncertain,and even can be artificially camouflaged.In order to solve this problem,this paper uses clustering algorithm to extract the suspected leakage source,and creates the map accordingly.In order to improve the real-time performance of plume detection robot,a hierarchical topology map is created.The main contents are as follows:1)In order to quickly create hierarchical topology map in real scene,firstly,feature extraction algorithm is used to quickly extract the feature points of each object from the environment,and then the suspected leakage source is extracted.Based on the research of classical feature extraction algorithms,such as sift,surf and so on,this paper selects orb algorithm which has the advantages of fast and BREF algorithm,and has good real-time and accuracy to extract and describe feature points,which lays a foundation for plume detection robot to quickly extract suspected leakage sources and map creation.2)In order to avoid the discrepancy between the shape and category of suspected leakage source and the actual situation,a clustering algorithm based on scale division and region growth(SRG clustering algorithm)is proposed,which is not affected by prior knowledge.On the one hand,the algorithm improves the accuracy of clustering,on the other hand,it reduces the number of sample points to a certain extent,and saves the time of region growth traversing sample points.On the one hand,it can identify clusters of arbitrary shape by region growing,on the other hand,it does not need to calculate the similarity between sample points,it only needs to traverse the sample points for clustering.Experiments on synthetic data sets and UCI data sets show that SRG clustering algorithm is better than other clustering algorithms in ACC,precision,recall,F1 metric and NMI.For example,in terms of F1 metrics combining precision and recall,SRG clustering algorithm is 39.8% higher than k-means algorithm,34.9% higher than FCM clustering algorithm,17.8% higher than DBSCAN algorithm,and 22.5% higher than DPC algorithm.3)According to the clustering results of the real scene,the hierarchical topology map is created,and the rationality of the topology nodes,the optimal path of the topology map and the shortest path of the whole detection range are analyzed.After comprehensive comparison,it is concluded that the hierarchical map based on SRG clustering algorithm has a clearer meaning of the topology nodes,neither over clustering nor over clustering On the other hand,in four real scenes,the optimal path of the topology map and the shortest path of the whole detection range are better than other algorithms.