分析基于航拍图像的病虫害识别场景,发现该场景对网络中浅层特征图中小目标的信息提取要求较高。为解决这个问题,对SSD网络进行了改进,优化了目标检测层,提出一种双主干网络的策略DB-SSD（Dual Backbone SSD）,即在保持原本的VGG16网络不变的基础上,额外增加一条更能表示浅层细节信息的ResNet50作为次主干网络,来丰富特征图的表达。实验表明,在识别基于航拍图像的病虫害方面,所提检测算法具有明显的优势。

Analyzing the scene of pest and disease recognition based on aerial images, it is found that this scene requires high information extraction of small targets in shallow feature maps in the network. To address this issue, improvements are made to the SSD network by optimizing the object detection layer and proposing a dual backbone network strategy called DB-SSD (Dual Backbone SSD). This strategy involves adding an additional ResNet50 as a secondary backbone network that can better represent shallow details while maintaining the original VGG16 network, in order to enrich the expression of feature maps. Experiments have shown that the proposed detection algorithm has significant advantages in identifying pests and diseases based on aerial images.