基于MobileNet V2的田间葡萄果穗成熟度判别

Grape ripeness discrimination based on MobileNet V2

葡萄的成熟度主要依靠人的感官判断,缺乏客观标准,严重影响葡萄的整体品质。针对以上问题,本文采用MobileNetV2轻量级的卷积神经网络模型对葡萄成熟度进行判别。田间采集红提葡萄图像共计380张,构建数据集,以不同着色度为指标,对数据集进行分类,并按7:2:1分为训练集、验证集、测试集,将训练集、验证集预处理后采用该数据集训练MobileNetV2模型。由测试集的试验结果,得到试验训练MobileNetV2模型所获得的分类准确率为87%。最后采用QT语言设计了可视化的界面。本文的研究将有利于推进葡萄采摘的自动化,降低葡萄采摘成本、提高葡萄的经济效益。

The maturity of grapes mainly depends on human sensory judgment, lack of objective standards, which seriously affects the overall quality of grapes. In response to the above problems, this paper uses the MobileNet V2 lightweight convolutional neural network model to discriminate the ripeness of grapes. A total of 380 images of red grapes were collected in the field, and a data set was constructed. The data set was classified according to different coloring degrees, and divided into training set, validation set and test set according to 7:2:1. The dataset is used to train the MobileNet V2 model after preprocessing.According to the experimental results of the test set, the classification accuracy obtained by training the MobileNet V2 model in this paper is 87%. Finally, the visual interface is designed using QT language. The research in this paper will help to promote the automation of grape picking, reduce the cost of grape picking and improve the economic benefits of grapes.