Dual Low Identification Target Recognition in Complex Environment based on Neural Network
Yuan Wu, Keke Geng, Peilin Xue, Guodong Yin, Wei Zou, Shuaipeng Liu, Yongjun Yan · 2019
In order to solve the problem that the traditional neural network has low recognition rate in the complex environment, this paper designed a low recognition target recognition system based on Faster-RCNN and YOLO neural network. Firstly, a synchronous acquisition system for collecting RGB and thermal dual datasets was designed. Then, a Dual Faster-RCNN and a Dual YOLO neural network were established, and three sets of information feature fusion experiments were carried out to establish the optimal network parameters. Finally, the two networks that were trained and tested using a separate test sets. The test data shows that the average recognition rate of personnel and vehicles in the Dual Faster-RCNN neural network is about 82%, and the recognition rate of the Dual YOLO neural network is about 80.3%. The low-recognition target recognition system studied in this paper can accurately identify the key information in low-recognition pictures, which has important research significance and practical application value for future development of intelligent transportation, remote sensing image analysis and security system protection.