The current problem of relatively backward and inefficient apple fruit grading technology, computer vision-based classification methods are widely adopted, but traditional visual classification networks face the problems of many parameters, high computational effort and unsatisfactory classification accuracy. Therefore, this paper proposes a lightweight residual network-based apple external quality grading method. Firstly, based on the traditional residual neural network, the network uses group convolution to replace the standard convolution in the original residual units, the aims are to reduce the number of model parameters and computational effort; Secondly, to address the information non-circulation problem between group channels caused by group convolution, a Channel Shuffle operation is used to mix inter-group features to improve model performance; Finally, a parallel pooling structure is proposed to solve the problem of information loss of traditional pooling features. To build a dataset of apple images with extensive coverage of external quality information, and to perform data enhancement on a limited dataset, and to conduct experiments based on the augmented dataset using improved models in comparison with common neural network models. The experimental results show that the improved lightweight residual network model has only 2.97M parameters, the FLOPs are only 1/5 of the traditional model, and the classification accuracy is 96.5%, which is helpful for future implementation of apple grading in low performance mobile terminals.
The inability to extract and utilize features in semantic segmentation leads to the loss of detailed information and the discontinuity of semantic information, and this study proposed an encoding and decoding model CAF-Deeplabv3+ for semantic segmentation based on coordinate attention and feature fusion. First, the coordinate attention module (CA) is embedded in the backbone network ResNet for the feature extraction; Secondly, a strip pooling branch (SP) is added to the ASPP module in encoding structure to help capture contextual information; In the process of feature reuse, a feature fusion module (FFM) for fusing low-level features and high-level features. It has been verified that the model can enhance the feature extraction capabilities. Our method achieved excellent results on Cityscapes. The training and validation on the Cityscapes dataset showed that the experimental results achieve 75.01% of the mean intersection over union (mIOU), which is improved by nearly 2%.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.