Nursery level bamboo plant species classification using Deep Learning

Abstract

Bamboos are of considerable value in farming and are considered as a multi-utility “Green Gold” resource. They also play a crucial role in supporting human life due to their environmental significance. Several bamboo species are difficult to classify manually using traditional methods. Due to the subtle differences in anatomy, their visual identification requires much field experience and expertise. This raises challenges for visual identification and suggests that the use of modern technologies could help with their identification. Accurate identification would assist in choosing the appropriate species for cultivation and would help ensure the appropriate species for various end-use applications. In this study, deep learning (DL) and machine learning (ML) were used for the automated identification of bamboos growing in nurseries. We created a bamboo image dataset comprising of 800 plant images and 4000 augmented images. These were categorized into four distinct species. We used transfer learning methods with and without image augmentation. The results revealed that DenseNet121, VGG16, VGG19 and MobileNetV2 could identify plants with an accuracy of 80.83 %, 82.92 %, 84.17 % and 76.67 %, respectively, without augmentation and with 83.37 %, 95.87 %, 96.37 % and 88.12 % accuracy, respectively, with augmentation. We introduced a Classifusion approach to classification which resulted in an accuracy rate of 97.37 % for augmented images. The results of this work should help farmers and novice users to identify species automatically at the nursery stage.