Visualization of Tree Species Identification Using Mask RCNNs for Tropical Forests in Malaysia

A variety of species support environmental efforts, and drones provide easier and more precise imaging than satellites in comparison to their satellite data due to their vastly improved image quality compared to satellite data. This project seeks to establish a novel method for recognizing tree species based on drone imagery that can be evaluated. As part of this project, we will design a novel model for identifying tree species using drone imagery, which we will then test and evaluate it. This project will gather aerial images of forests using a drone. After aerial images have been obtained, a bespoke dataset based on the captured aerial images will be developed. To estimate the tree species based on aerial images from this dataset, we will use machine learning to make predictions using this information. This research initiative intends to collect photographs of Malaysian forests and their fauna, as Malaysian species are distinct from those of other nations due to the country's climate. The purpose of this study is to identify and segment tree species from high-resolution RGB images of tropical forests using the Mask R-CNN method, a robust system for detecting tree species. The developed algorithm was validated on two datasets in which the experimental results indicate a mean average precision (mAP) of 19.17%, 67.9% for precision, and 16.9% for recall for the improved model. Specifically, mAP improved by 6.03%, 46.80% precision, and 2.27 recall over the original Mask R-CNN. Using different datasets, the enhanced model demonstrated high detection and segmentation accuracy as well as making it suitable for providing technical support in intelligent forest management. A successful outcome of this endeavor could result in the first phase of a forest inventory, which would have repercussions for logging and forest management in the region.