基于MobileNet的果蔬识别系统
2021-02-19陈怡帆






摘 要:我国疆域辽阔,土壤肥沃,气候温和,尤其是新疆地区,日照时間充足,盛产水果。造就了我国成为农业生产大国,每年进出口非常多的水果蔬菜。据了解,在大部分农贸市场都靠人工进行果蔬分类,工作量多且效率低下。提出一种基于MobileNet模型的果蔬识别系统,该系统可以快速进行果蔬识别。该项目用了传统CNN模型和更轻量化的MobileNet模型对12个不同品种的蔬果数据集进行训练,发现基于MobileNet模型的识别结果正确率更高。
关键词:深度学习;图像识别;卷积神经网络;MobileNet
中图分类号:TP391.4 文献标识码:A文章编号:2096-4706(2021)13-0155-04
Fruit and Vegetable Recognition System Based on MobileNet
CHEN Yifan
(School of Computer and Software, Jincheng College of Sichuan University, Chengdu, 611731, China)
Abstract: China has a vast territory, fertile soil and mild climate. Especially in Xinjiang, it has sufficient sunshine and is rich in fruits. As a result, China has become a large agricultural production country, importing and exporting a lot of fruits and vegetables every year. It is understood that in most farmers’ markets, fruits and vegetables are classified manually, which has a large of workload and low efficiency. Therefore, this paper puts forward a fruit and vegetable recognition system based on MobileNet model, which can quickly recognize fruits and vegetables. The project uses the traditional CNN model and the lighter MobileNet model to train the data sets of 12 different kinds of fruit and vegetable. It is found that the recognition result based on MobileNet model has a higher accuracy.
Keywords: deep learning; image recognition; convolutional neural network; MobileNet
0 引 言
近年来,随着社会的快速发展,经济贸易的逐渐扩大,我们国家已经成为了一个商品进出口大国,特别是农业商品,每年的进出口量非常之大。在调查过程中,发现国内大部分农贸市场并没有涉及有水果蔬菜识别领域的系统,大多数地方都采用人工工作,于是本实验项目设计了一个果蔬识别系统,预期可以用于果蔬识别及分类,有效提高农贸市场工作效率。该设计采用了卷积神经网络(CNN),CNN已经普遍应用在计算机视觉、自然语言处理领域,并且已经取得了不错的效果。所以在图像识别领域,必然离不开广泛使用的卷积神经网络。本项目使用了传统CNN和轻量级MobileNet网络模型,并借此调用数据集进行训练,再对训练后的两个模型进行测试,并对它们的结果进行对比。
1 MobileNet网络
1.1 什么是MobileNet网络
MobileNet网络是专注于移动端或者嵌入式设备中的轻量级CNN网络,相比传统神经网络,在准确率小幅降低的前提下大大减少了模型参数和运算量,具体可以在下面公式的推算中证明。……
