摘要
为了提高机械臂采摘的效率及增强安全自主性,提出了一种基于改进卷积神经网络和迁移学习的计算机图像识别模型。首先,建立了由可采摘与不可采摘图像组成的样本数据集,将每幅图像的像素设置为256×256;然后,构建基于改进卷积神经网络和迁移学习的计算机图像识别模型,并将自动编码机网络结构与卷积神经网络运算方法相结合,利用自动编码机网络结构具有编码和解码的环节,通过卷积神经网络运算方式构建出一种改进的卷积神经网络;通过卷积层挖掘图片信息中具有采摘信息的特征,同时消除随机环境对图片的干扰,解码部分能够对特征图像进行上采样并判断是否应该进行采摘与采摘姿势;最后,将构建网络模型与迁移学习相结合进行实验,分析迁移学习方法、数据集样本大小、网络参数对实验结果的影响。结果表明:采摘机械臂识别模型整体识别率更高,能够构建出效率更高、鲁棒性更强的采摘控制系统。
In order to improve the efficiency of robot arm picking and enhance safety autonomy,this paper proposes a computer image recognition model based on improved convolutional neural network and transfer learning.First,establish a sample data set consisting of the collectable and non collectable images,and set the pixel of each image to 256×256;Then,a computer image recognition model based on improved convolutional neural network and transfer learning is constructed.This model combines the network structure of the automatic encoder with the convolutional neural network operation method.Using the network structure of the automatic encoder with the link of encoding and decoding,an improved convolutional neural network is constructed through the convolutional neural network operation mode.The convolution layer is used to mine the features with picking information in the picture information,and at the same time,the interference of the random environment on the picture is eliminated.The decoding part can up sample the feature image and judge whether to pick and pick the pose;Finally,we combine the construction of network model with migration learning to conduct experiments,and analyze the impact of migration learning methods,data set sample size,and network parameters on the experimental results.The experimental results show that the overall recognition rate of the recognition model proposed in this paper is higher.This method can build a harvesting control system with higher efficiency and stronger robustness.
作者
马琰
Ma Yan(Wuxi Vocational Institute of Arts&Technology,Yixing 214200,China)
出处
《农机化研究》
北大核心
2024年第12期208-212,共5页
Journal of Agricultural Mechanization Research
基金
江苏省教育科学研究院十三五规划项目(2018-R-66945)。
关键词
采摘机械臂
计算机视觉
卷积神经网络
迁移学习
picking mechanical arm
convolution neural network
computer vision
transfer learning
