基于PP-YOLO的农业病虫害识别算法

Recognition Algorithm of Agricultural Diseases and Insect Pests Based on PP-YOLO

为解决因害虫尺度多样性导致其识别度相对较低的问题,本研究提出了一种基于PP-YOLO(PaddlePaddleYou Only Look once)的农业病虫害识别算法。选取2359个病虫害样本数据集,按照9∶1的比例进行训练集、测试集的划分;选择PP-YOLO模型进行病虫害监测,并利用平均精度mAP(mean average precision)指标进行模型精度评价;探讨PP-YOLO结合数据增强mixup、颜色扭曲法在病虫害中小目标检测上的适用性。结果表明,PP-YOLO模型在病虫害中小目标检测方面mAP达47.4%、26.5%;基于PP-YOLO模型结合数据增强mixup与颜色扭曲后在病虫害中小目标检测上mAP分别提升4.3%、2.9%。总之,PP-YOLO模型可有效检测识别农作物害虫,同时,数据增强mixup与颜色扭曲法可有效提升病虫害的数据样本指标。

In order to solve the problem that the recognition degree of pests is relatively low due to the scale diversity of pests,this study proposed an agricultural pest identification algorithm based on PP-YOLO.A total of 2359 sample data sets were selected,and the training set and test set were divided according to the ratio of 9∶1.The PP-YOLO model was selected for pest detection,and the model accuracy was evaluated by using map index.The small and medium-sized objectives of the method of PP-YOLO combined with the data enhancement mixup and color distortion were discussed applicability of detection.The map of the PP-YOLO model was 47.4%and 26.5%in the detection of small and medium-sized targets of diseases and insect pests.Based on the PP-YOLO model,the map was increased by 4.3%and 2.9%respectively after the combination of data enhancement mixup and color distortion.The PP-YOLO model proposed in this paper could effectively detect and identify crop pests.At the same time,data enhancement mixup and color distortion could effectively improve the data sample index of pests and diseases.