An Improved CNN-Based Apple Appearance Quality Classification Method With Small Samples

Li Sun

doi:10.1109/ACCESS.2021.3077567

2021-01-01

DOI：10.1109/ACCESS.2021.3077567

原文链接

查看原文

摘要

En 中文

Apple quality classification is an important means to refine apple sales market and promote apple sales. At present, most of classification methods based on a convolutional neural network (CNN) depend on the quantity of training samples to get good performance. But due to the lack of large-scale public apple appearance dataset, it is a big challenge to obtain high accuracy of apple appearance quality classification with small samples. Therefore, we propose an improved method based on CNN for apple appearance, quality classification with small samples. Firstly, support vector machine (SVM) is used for image segmentation to avoid the decrease of recognition accuracy caused by environmental noise. Secondly, the segmented image data are input into deep convolutional generative adversarial networks (DCGAN) model, which is used for data expansion. Thirdly, the improved ResNet50 (Imp-ResNet50) is proposed as follows: Replace the fully-connected layer with global average pooling layer; Add the dropout algorithm and batch normalization algorithm at the fully-connected layer; Replace the activation function ReLU with Swish. Through comparative experiments with 360 apple images, we verify the performance of the proposed method including the training image quality, the running time, and classification accuracy. The result shows that the proposed method can obtain high quality training samples and reduce the running time of the method effectively. At the same time, it can realize higher classification accuracy that is up to 96.5%, which is higher than the previous classification method.

Keyword:

Support vector machines

Image segmentation

Data models

Feature extraction

Training

Mathematical model

Image recognition

Apple quality classification

SVM

DCGAN

ResNet50