基于Faster R-CNN的新疆棉花幼苗与杂草识别方法

Identification method of cotton seedlings and weeds in Xinjiang based on Faster R-CNN

针对新疆棉田杂草的伴生特点带来的特征过拟合、精确率低等问题,以新疆棉花幼苗与杂草为研究对象,分析杂草识别率低的影响因素,建立了基于Faster R-CNN的网络识别模型.采集不同角度、不同自然环境和不同密集程度混合生长的棉花幼苗与杂草图像5370张.为确保样本质量以及多样性,利用颜色迁移和数据增强来提高图像的颜色特征与扩大样本量,以PASCAL VOC格式数据集进行网络模型训练.通过综合对比VGG16,VGG19,ResNet50和ResNet101这4种网络的识别时间与精度,选择VGG16网络训练Faster R-CNN模型.在此基础上设计了纵横比为1∶1的最佳锚尺度,在该模型下对新疆棉花幼苗与杂草进行识别,实现91.49%的平均识别精度,平均识别时间262 ms.研究结果为农业智能精确除草装备的研发提供了参考.

Aiming at the problems of over-fitting and low accuracy caused by the accompanying cha-racteristics of weeds in cotton fields in Xinjiang,taking cotton seedlings and weeds in Xinjiang as research objects,the factors affecting the low recognition rate of weeds were analyzed,and a network recognition model based on Faster R-CNN was established.A total of 5370 cotton seedlings and weeds were collected from different angles,different natural environments and different intensive levels.In order to ensure sample quality and diversity,color migration and data enhancement were used to improve the color characteristics of the image and to expand the sample size.In addition,the network model training was performed in the PASCAL VOC format data set.By comprehensively comparing the recognition time and accuracy of the four networks VGG16,VGG19,ResNet50 and ResNet101,the VGG16 network training Faster R-CNN model was selected.On this basis,the optimal anchor scale with aspect ratio of 1∶1 was designed.Based on this model,Xinjiang cotton seedlings and weeds were identified,and the average recognition accuracy of 91.49%was achieved.The average recognition time was 262 ms.The results provide a reference for the development of agricultural intelligent precision weeding equipment.