Automated Identification of Tree Species by Bark Texture Classification Using Convolutional Neural Networks

Abstract

Identification of tree species plays a key role in forestry related tasks like forest conservation, disease diagnosis and plant production. There had been a debate regarding the part of the tree to be used for differentiation, whether it should be leaves, fruits, flowers or bark. Studies have proven that bark is of utmost importance as it will be present despite seasonal variations and provides a characteristic identity to a tree by variations in the structure. In this paper, a deep learning based approach is presented by leveraging the method of computer vision to classify 50 tree species, on the basis of bark texture using the BarkVN-50 dataset. This is the maximum number of trees being considered for bark classification till now. A convolutional neural network(CNN), ResNet101 has been implemented using transfer-learning based technique of fine tuning to maximise the model performance. The model produced an overall accuracy of >94% during the evaluation. The performance validation has been done using K-Fold Cross Validation and by testing on unseen data collected from the Internet, this proved the model's generalization capability for real-world uses.

MikrobatX: The Use of SIFT Feature Extraction in a Deep Learning Approach for Identification and Classification of Microscopic Fragments of Medicinal Leaves

“…Artificial intelligence (AI) has been trending in the research community for the past few years. Researchers and companies widely use artificial intelligence, machine learning, and deep learning in digital image processing by researchers and companies, for example, for plant identification and classification [34]- [38]. Previous research has carried out several recognition techniques using shape, size, texture, and descriptors.…”

Improvement and Assessment of Convolutional Neural Network for Tree Species Identification Based on Bark Characteristics

“…Robert et al developed DeepBark, which is a model capable of detecting bark surfaces under high background brightness [25]. Faizal achieved promising results on BarkVN-50 using a deeper network called ResNet101 [26]. and EfficientNet, obtained an identification accuracy value above 90%, and applied class activation mapping (CAM) aggregation to identify the critical classification features for each tree species [27].…”

“…Most existing studies on tree species identification use pre-trained weights of networks that were trained on ImageNet, rather than using bark images as the pre-training data [22][23][24][25][26][27]. This method could lead to some misclassification and performance degradation.…”

“…This method could lead to some misclassification and performance degradation. Many researchers use relatively backward networks to work on image classification of tree species [22][23][24][25][26] (like ResNet). Compared to some algorithms developed in recent years, the performance of traditional convolutional neural networks is inferior.…”