基于改进SSD的苹果叶部病理检测识别

Apple leaf pathology detection and recognition based on improved SSD

针对目前主流的目标检测算法在苹果叶部病理的检测中识别速度和精度较低的问题,实现了基于改进SSD的苹果叶部病理的检测识别。首先,采用轻量级特征融合结构,融合高低层特征图特征;其次,引入通道注意力机制,提取更有效的病斑小目标特征信息,同时使用Focal Loss损失函数代替原有的Multibox Loss损失函数,减少了训练中大量简单负样本的权值;最后,利用苹果叶部病理公共数据集进行对比实验,选取训练最优的网络。实验表明:改进的SSD比其它算法的检测效果有明显的提升。

Aiming at the problem of low recognition speed and precision of the current mainstream target detection algorithms in detection of apple leaf pathology,the detection and recognition of apple leaf pathology based on improved single shot multibox detector(SSD)is realized.Firstly,lightweight feature fusion structure is adopted to fuse features of high and low layer feature map.Secondly,the channel attention mechanism is introduced to extract more effective feature information of small target of disease spots.At the same time,Focal Loss function is used to replace the original Multibox Loss function,which reduces the weight of a large number of simple negative samples in training.Finally,the public dataset of apple leaf pathology is used for comparison experiment,and the optimal training network is selected.The experiment shows that the improved SSD significantly improves the detection effect compared with other algorithms.