摘要
为解决田间自然光照条件下绿芦笋自主采收中的作物识别问题,提出了一种将图像预处理与CNN算法相结合的方法。对于获取的原始图像,首先在Lab和YUV颜色空间下的a通道和U通道进行OTSU阈值分割,之后合并分割图并进行去噪处理,简化突出图像特征,再根据预处理后图像特点,优化改进LeNet网络结构,构建CNN模型对预处理后图像进行识别,提取目标作物绿芦笋。试验结果表明:本方法可实现田间自然光照条件下绿芦笋的有效识别,识别准确率为89.39%,可为后续绿芦笋自主采收设备的研究奠定基础。
In order to solve the difficult problem of green asparagus recognition under field illumination conditions in the process of autonomous harvesting,an approach combining the image preprocessing and CNN algorithm was proposed in this article.The original image acquired in the field environment was firstly transformed into Lab and YUV color space and segmented by OTSU method in‘a’channel and‘U’channel respectively.Then the two segmented images obtained were merged to get a better final segmented image,and the operation of noise removal was performed.After image preprocessing,the features of the original image could be simplified and highlighted.Then according to the features of preprocessed image,the structure of LeNet network was improved to establish a better CNN model to recognize the preprocessed image and extract the target green asparagus.The testing results showed that the method proposed in this paper can effectively recognize the green asparagus under field illumination conditions,and the accuracy reached 89.39%.Therefore,the method in this paper can provide technical basis for the research of green asparagus autonomous harvesting device in the future.
作者
李鹏
刘翔鹏
李彦明
刘成良
Li Peng;Liu Xiangpeng;Li Yanming;Liu Chengliang(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《农机化研究》
北大核心
2021年第7期19-25,共7页
Journal of Agricultural Mechanization Research
基金
上海市青年科技英才扬帆计划项目(18YFl411000)
上海市科技兴农项目(2019-02-08-00-08-F01122)。
关键词
视觉识别
图像处理
阈值分割
CNN
绿芦笋
visual recognition
image processing
threshold segmentation
CNN
green asparagus
