Flower Species Recognition System Combining Object Detection and Attention Mechanism

Abstract

In this paper, a flower species recognition system combining object detection and Attention Mechanism is proposed. In order to strengthen the ability of the model to process images under complex backgrounds, we apply the method of object detection to locate flowers that we want to recognize. For less time of training, object detection and classification are Integrate into an end-to-end network stacked attention modules to generate attention-aware features. Experiments are conducted on Flower 102, our method can recognize flower species against a complex background. With model owning attention module and transfer learning, we increase mAP from 73.8% to 74.7% and training time is reduced by about 15%.