基于双目图像深度学习的农作物择优采摘仿真
Simulation of Crop Picking Based on Binocular Image Depth Learning
白维维 1李俊杰 1陈烽2
作者信息
- 1. 凯里学院微电子与人工智能学院,贵州 凯里 556100
- 2. 西藏民族大学信息工程学院,陕西 咸阳 712082
- 折叠
摘要
面对复杂的农作物生长环境,利用传统机器视觉技术采摘易受到未成熟果实以及周围叶片影响,获取的农作物果实图像存在较多的不可用信息,无法完成择优采摘.为了确保采摘的果蔬品质,提出基于双目图像深度学习的农作物择优采摘方法.利用直方图均衡化变换视觉图像区域,明确农作物图像内像素灰度值,均衡化色块不均位置.通过色彩分量调节全局图像颜色,以颜色差异分割双目图像,剔除局部RGB色彩关联性.在提取农作物形状特征前,将深度信息再次归一化,获得作物形态描述符.选择卷积神经网络对图像实行卷积运算,将择优特征结果输入到卷积层内,输出图像分类结果,实现农作物择优采摘.实验结果表明,所提方法的择优采摘精准度达到0.98,果实形状特征识别为 0.96.说明所提方法能够准确识别出品质佳的农作物,实现了择优采摘.
Abstract
In order to ensure the quality of fruits and vegetables,based on binocular image deep learning,this pa-per presented a method of picking the best crop.Firstly,histogram equalization was used to transform the visual image region,and then the gray values of pixels in the crop image were determined.Meanwhile,the uneven position of color blocks was equalized.Secondly,global image color was adjusted by color components,and then the binocular image was segmented by color difference.Moreover,local RGB color correlation was eliminated.Before extracting crop shape features,the depth information was normalized again,so that the crop shape descriptor could be obtained.Furthermore,the convolutional neural network was used for convolution operation on the image.Finally,the results of selecting features were input into the convolutional layer,and then image classification results were output.Thus,the selective picking of crops.Experimental results show that the best picking accuracy of the proposed method can reach 0.98,and the recognition rate of fruit shape characteristics is 0.96,indicating that the proposed method can accurately identify the crops with good quality and realize optimal picking.
关键词
双目图像/深度学习/农作物择优采摘/图像预处理/特征提取Key words
Binocular image/Deep learning/Pick the best crops/Image preprocessing/Feature extraction引用本文复制引用
基金项目
黔东南州基础研究计划(2022)(黔东南科合基础[2022]11号)
扶持市(州)高校质量提升工程项目(2022)(30)
出版年
2024
NETL
NSTL
万方数据