摘要
针对当前大田环境条件下对害虫进行识别研究的不足,以南方蔬菜重大害虫为研究对象,探索了一种在大田环境下使用黄色诱捕板对蔬菜害虫进行监测计数的方法。在经典图像处理算法基础上,根据害虫监测目标的需要,提出了一种基于结构化随机森林的害虫图像分割算法和利用不规则结构的特征提取算法,进一步结合背景去除、干扰目标去除和检测模型计数子算法,集成设计了基于视觉感知的蔬菜害虫计数算法(Vegetable pest counting algorithm based on visual perception,VPCA-VP)。使用了现场环境下拍摄的图像进行实验与分析,共识别出蓟马9351只,烟粉虱202只,实蝇23只。经过与人工计数比对得出,本文基于视觉感知的蔬菜害虫计数算法的平均识别正确率为94.89%。其中,蔬菜害虫蓟马的识别正确率为93.19%,烟粉虱的识别正确率为91%,实蝇的识别正确率达到100%。算法达到了较好的测试性能,可以满足害虫快速计数需求,在农田害虫监测中有一定的应用前景。
Due to the varying degree of various pests ＇ damage,people tend to make some counter measures to protect the vegetables. Up to now,the most common method is to spray pesticides on vegetable pests. Farmers often lead to the excessive use of pesticides for lack of information about the number of pests. Traditionally,manual counting methods are carried out on the number of pests. It needs large labor costs,heavy workload,with subjective and other shortcomings,and using machine vision to monitor vegetable pests is a popular method recently. But the vast majority of current visual methods are to be carried out under the condition of ideal laboratory,which cannot be directly applied to pest monitoring in the field. Using visual perception technology to identify pests has become a hotspot in the field of agricultural engineering in recent years. Because of the shortcomings of the pests identification under the current field conditions,a new algorithm for counting the southern vegetable pests was studied by using yellow sticky trap. Based on the classical image processing algorithm,some new algorithms,including pest image segmentation sub-algorithm based on the structure of random forest, feature extraction sub-algorithm of irregular structure,background removal sub-algorithm,interference target removal sub-algorithm and detection model counting sub-algorithm were proposed. Those sub-algorithms were integrated to create a vegetable pest count algorithm based on visual perception（ VPCA-VP）. The images taken in the field environment were used for experimentation and analysis,and 9351 thrips,202 whiteflies and 23 fruit flies were recognized. Compared with the artificial count,the accuracy rate of the vegetable pest counting algorithm based on visual perception was 94. 89%. Among them,the accuracy rate of the thrip was 93. 19%,the accuracy rate of the whitefly was 91% and the exact rate of the fruit flywas 100%. The algorithm had good performance and achieved the rapid counting demand,which had wide application prospec
作者
肖德琴
张玉康
范梅红
潘春华
叶耀文
蔡家豪
XIAO Deqin1,2 ZHANG Yukang1 FAN Meihong3 PAN Chunhua1 YE Yaowen1 CAI Jiahao 1(1. College of Mathematics and Informatics, South China Agricultural University, Guangzhou 510642, China 2. Guangdong Province Agricultural Data Engineering Research Center, Guangzhou 510642, China 3. Guangzhou Golden Farm Co. , Ltd. , Guangzhou 511470, Chin)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2018年第3期51-58,共8页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家星火计划项目(2015GA780002)
广东省科技计划项目(2016B010110005
2015A020209153
091721301064071007)
关键词
视觉感知
蔬菜害虫
识别
随机森林
相似性描述子
不规则特征提取
visual perception
vegetable pest
identification
random forest
similarity descriptor
irregular feature extraction
