Research Article
Apple Grading Model Based on Improved ResNet-50 Network
@INPROCEEDINGS{10.1007/978-3-031-18123-8_59, author={Lei Zhao and Qinjun Zhao and Tao shen and Shuhui Bi}, title={Apple Grading Model Based on Improved ResNet-50 Network}, proceedings={Multimedia Technology and Enhanced Learning. 4th EAI International Conference, ICMTEL 2022, Virtual Event, April 15-16, 2022, Proceedings}, proceedings_a={ICMTEL}, year={2022}, month={10}, keywords={Apple grading ResNet network Attention mechanism LeakyReLU activate function}, doi={10.1007/978-3-031-18123-8_59} }- Lei Zhao
Qinjun Zhao
Tao shen
Shuhui Bi
Year: 2022
Apple Grading Model Based on Improved ResNet-50 Network
ICMTEL
Springer
DOI: 10.1007/978-3-031-18123-8_59
Abstract
In this paper, we study an apple grading model based on the convolutional neural network to classify Red Fuji apples according to features of size, color and external defects. Firstly, Red Fuji apple images are collected by professional equipment, and the RGB model of apple image is extracted and transformed into HSI model. Secondly, the segmentation between apple and background is realized by Otsu method in the S channel. Thirdly, the ResNet-50 network is improved by convolutional block attention module and LeakyReLU activation function. Finally, improved ResNet-50 network is applied to apple grading and compared with other mainstream convolutional neural networks. The experimental result shows that improved ResNet-50 network reaches the highest accuracy 95.1% in apple grading experiment, which is higher than AlexNet, VGG-16, GoogleNet, Mobilenet-V2 and the ResNet-50 network.
