基于YOLOv5的害虫识别研究

Pest Identification Based on YOLOv5

针对害虫图像数据中小目标误检、漏检、类别不平衡及特征提取能力不足等问题,提出了一种改进的基于YOLOv5的害虫检测模型。首先,该算法通过伪标签技术缓解标签数量不足带来的问题;其次,通过增加一个4倍的下采样层,调整损失函数增强对少数类别的感知能力;再次,通过修改目标框回归公式解决训练过程中梯度消失的问题,提升小目标的检测精度;最后,利用虫情测报灯采集的图像数据进行实例验证。实验结果表明,该害虫检测模型具有较好的预测效果。

To solve the problems of false detection,missed detection,category imbalance and insufficient feature extraction ability of small targets in pest image data,an improved pest detection model based on YOLOv5 is proposed.Firstly,the algorithm uses pseudo-label technique to deal with insufficient labels.Secondly,a 4-fold down-sampling layer and focal loss function are introduced to catch minority categories better.Thirdly,the target box regression formula is modified to solve the problem of gradient disappearance in the training process and improves the accuracy of small targets detection.Finally,the image data collected by the pest detection lamp is used for experiment.The experimental results show that the pest detection model has a good pr ediction effect.