Identification and recognization of bamboo based on cross-sectional images using computer vision

Identification of bamboo is of great importance to its conservation and uses. However, identify bamboo manually is complicated, expensive, and time-consuming. Here, we analyze the most evident and characteristic anatomical elements of cross section images, that's a particularly vital breakthrough point. Meanwhile, we present a novel approach with respect to the automatic identification of bamboo on the basis of the cross-sectional images through computer vision. Two diverse transfer learning strategies were applied for the learning process, namely fine-tuning with fully connected layers and all layers, the results indicated that fine-tuning with all layers being trained with the dataset consisting of cross-sectional images of bamboo is an effective tool to identify and recognize intergeneric bamboo, 100% accuracy on the training dataset was achieved while 98.7% accuracy was output on the testing dataset, suggesting the proposed method is quite effective and feasible, it's beneficial to identify bamboo and protect bamboo in coutilization. More collection of bamboo species in the dataset in the near future might make EfficientNet more promising for identifying bamboo.