Research on Target Recognition Method Based on Integrated Learning

摘要

In order to improve the accuracy of obstacle recognition in front of vehicles, a method of obstacle recognition based on heterogeneous integrated learning is proposed. The heterogeneous data includes video image and radar text data. The purpose of heterogeneous integration is to increase the attention of vulnerable groups. In this method, alexnet, vggnet-16 and text CNN are used to train models on large-scale datasets, and the migration learning method is used to migrate to the traffic obstacle datasets to train different classification models. Finally, the weighted voting method is used to build the integration model. The experimental results show that the accuracy of this method is 99.10% compared with the single recognition method, and it has a better recognition effect.