Single shot multibox detector for honeybee detection

Jintong Cai, Yugo Makita, Yuchao Zheng, Shiya Takahashi, Weiyu Hao, Yoshihisa Nakatoh

Abstract: In recent years, deep learning has made great strides in automatic feature extraction through learning. Our target is to operate computers to analyze honeybee behavior and facilitate computational behavior in biology through object detection and reveal a wealth of previously unknown information. Although the object detection task has been well developed, there is no relevant bee dataset. Therefore, in this study, we first created a honeybee dataset using manual annotation and applied data augmentation to it. Furthermore, this study trains a single-shot multiple-box detector (SSD) for object detection to detect honeybee regions using a manually created dataset. We check the accuracy of the proposed method using the mean percentage of fit (mAP) commonly used in SSD. The experimental results show that our method works well.