Growing period classification of Gynura bicolor DC using GL-CNN

Abstract

Determining the optimal growth stage of leafy vegetables suitable for harvesting is of great importance. Despite the availability of many intelligent programs designed for plant classification, identifying the growth stage of a vegetable from its leaf features remains a major challenge. Gynura bicolor DC (G. bicolor) is an important vegetable, and its leaves are harvested for culinary use at a specific growth stage. The produce quality of leaves is compromised when they are harvested at improper stage. A classification model named GL-CNN was proposed based on convolutional neural networks to address this issue. The proposed model merges the features using a network fusion strategy to expand the feature representation on the basis of the intact leaf and leaf patch image sets. The networks were validated using a new dataset of G. bicolor planted and collected by ourselves. “Early fusion” and “late fusion” networks were designed and compared with GL-CNN to verify the rationality of network fusion location. The test accuracy of GL-CNN reaches 95.63%, which is the best in the classification task.