Detection of Eupatorium Adenophorum Based on Migration Learning

Abstract

Eupatorium adenophorum is one of the most typical examples of invasive alien species in China. Invasion of Eupatorium adenophorum causes serious damage to ecological environment and affects economic development of agroforestry. As the key step in the entire prevention and treatment process, the detection of Eupatorium adenophorum is beneficial to the effective implementation of control measures. Therefore, this paper uses the improved YOLOv3 network to detect Eupatorium adenophorum. Data augmentation and migration learning methods are used to avoid overfitting problems in the model and improve robustness and generalization capabilities. Experimental results show that the average precision value of Eupatorium adenophorum test reached 54.22%. The speed and precision of test are slightly improved compared with the original network. The way of this paper can realize effective detection of Eupatorium adenophorum.