基于SSD网络模型改进的水稻害虫识别方法  

Improved Rice Pest Recognition Based on SSD Network Model

作　　者：佘颢 吴伶 单鲁泉 SHE Hao;WU Ling;SHAN Luquan(School of Information Science and Intelligent Technology, Hunan Agricultural University, Changsha 410128, China)

机构地区：[1]湖南农业大学信息与智能科学技术学院,湖南长沙410128

出　　处：《郑州大学学报（理学版）》2020年第3期49-54,共6页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(61101235)。

摘　　要：针对目前主流的目标检测算法在虫害监控系统中识别速度慢、准确度低的问题,提出一种基于SSD网络模型的水稻害虫识别方法。该算法用表征能力更强的特征金字塔代替SSD原有的多尺度特征图,同时改进了归一化和激活函数,使得模型对小目标的识别率更高、收敛性更好,从而提高了水稻害虫的识别率与检测速度。实验表明,相比于faster R-CNN算法,基于SSD改进的水稻虫害识别方法的mAP最高提升了6.6%,其识别速度最高提升8倍。At present,most of the target detection algorithms based on candidate box were applied to pest monitoring system.The identification of algorithms was slow and inaccurate.The method of rice pest identification based on SSD network model was proposed.The original multi-scale feature mapping of SSD was replaced by a feature pyramid with strong representation ability.The improved normalization and activation methods were used to improve the accuracy and convergence.Experiments showed the mAP of the improved SSD was 6.6%higher,and the recognition speed was 8 times faster than faster R-CNN.

关 键 词：SSD神经网络 目标检测 数据增强 激活函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]