摘要
针对苹果叶部病害图像存在光照分布不均匀、对比度低、过亮或过暗区域细节丢失等问题,提出一种改进的Faster R-CNN苹果叶部病害检测方法,提高病害检测的准确率。由于HSV颜色空间中的H、S、V三个分量具有相对独立性,且光照及阴影部分的遮挡对H、S分量的影响很小,因此,将病害图像从RGB颜色空间转换到HSV颜色空间,再采用颜色恒常性(Retinex)算法对图像进行处理。然后,采用Faster R-CNN网络模型对苹果叶部的三种病害(雪松锈病、灰斑病、黑星病)进行目标检测。实验结果表明:该方法提升了检测苹果雪松锈病、灰斑病、黑星病的平均精度,分别提高了4.03%、7.14%和13.77%,整体平均精度提升了8.32%。每幅图像的检测时间为0.201 s,单张图片检测时间减少了42 ms,确保了检测的实时性,这对于病害的预防具有重要意义。
Aiming at the problems of uneven light distribution,low contrast,loss of detail in too bright or too dark areas in the image of apple leaf diseases,an improved apple leaf disease detection method based on Faster R-CNN is proposed to improve the accuracy of disease detection.Since the three components of H,S,and V in the HSV color space are relatively independent,and the occlusion of light and shadows has little effect on the H and S components,the RGB color space of the disease image is transformed to HSV color space.Then,the image is processed by the color constancy(Retinex)algorithm,which adaptively enhances the diseased image.It can not only accurately estimate the light component,but also improve the processing speed.Finally,the Faster R-CNN network model is used to detect the three diseases of apple leaves(cedar rust,gray spot,and scab).Experimental results show that this method improves the average accuracy of detecting apple cedar rust,gray spot,and black spot,respectively,by 4.03%,7.14%and 13.77%,and the overall average accuracy is increased by 8.32%.The detection time of each image is 0.201 s,a reduction of 42 ms,which ensures the real-time detection.This research is of great significance for the diseases prevention.
作者
王远志
施子珍
张艳红
WANG Yuanzhi;SHI Zizhen;ZHANG Yanhong(School of Computer and Information,Anqing Normal University,Anqing 246133,China)
出处
《安庆师范大学学报（自然科学版）》
2022年第2期26-30,共5页
Journal of Anqing Normal University(Natural Science Edition)
基金
安徽省教育厅自然科学研究重点项目(KJ2018A0359)
国家重点研发计划项目(SQ2020YFF0402315)。
作者简介
王远志(1977-),男,安徽宿松人,安庆师范大学计算机与信息学院教授,研究方向为数据库、图形学。E-mail:wangyuanzhi1@sohu.com。
