A Vehicle Model Data Classification Algorithm Based on Hierarchy Clustering

With wide application of deep learning in security field, using it on vehicle brand, style and years recognition product has become an active research. Due to the variety of vehicle brand, the total quantity of training samples needed by deep learning is so huge that the difficulty of sample collection and corresponding cost on time and labor are both unacceptable. In addition, new vehicle types come out continuously which require database augmentation and product update in time. To solve this problem, this article proposes a vehicle model data classification algorithm based on hierarchy clustering. Firstly, train the classification model with vehicle data collected by the index of vehicle model information. Secondly, get mean feature of each class and use hierarchical clustering according to the distance between the classes. Then on the basis of distance sorting and model test result to merge the vehicle models. Finally, the feasibility of this algorithm is verified through the experiment. Experimental results show the scheme is feasible. The algorithm realizes the automatic clustering of vehicle model data whose car face or tail has the same structure which can’t be distinguish in image or video. This article provides a new way for the development of vehicle brand, style and year recognition products.