Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (11): 214-218.DOI: 10.3778/j.issn.1002-8331.1906-0350

Previous Articles     Next Articles

Research on Granary Pest Detection Based on SSD

DENG Zhuanglai, WANG Pan, SONG Xuehua, WANG Changda, CHEN Juan, WU Liya   

  1. 1.School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China
    2.Joyea Co., Ltd., Danyang, Jiangsu 212300, China
  • Online:2020-06-01 Published:2020-06-01

基于SSD的粮仓害虫检测研究

邓壮来,汪盼,宋雪桦,王昌达,陈娟,吴立亚   

  1. 1.江苏大学 计算机科学与通信工程学院,江苏 镇江 212013
    2.江苏仅一联合智造有限公司,江苏 丹阳 212300

Abstract:

In order to effectively detect the pests in granary and reduce the loss of grain, this paper presents a method of pest detection for grain silos based on SSD. Multiple convolution feature maps are used to detect pests in this method. The training speed and detection efficiency of SSD are improved by lightweight model structure and loss function optimization. In this experiment, 6 kinds of high-outbreak granary pest images are trained and tested, and the results show that this method has a higher mAP in the detection of granary pests compared with the current mainstream object detection method.

Key words: granary pest, loss of grain, object detection, Single Shot MultiBox Detector(SSD), mAP

摘要:

为了对粮仓害虫进行有效地检测,减少粮食损失,提出一种基于SSD的粮仓害虫检测方法。该方法利用多个尺度的卷积特征图来检测害虫。通过轻量化模型结构和优化损失函数来提高SSD的训练速度和检测效率。实验将6类高爆发的粮仓害虫图像进行训练和测试,结果表明:该方法相比较于当前主流的目标检测方法在对粮仓害虫检测中具有更高的mAP。

关键词: 粮仓害虫, 粮食损失, 目标检测, SSD, mAP