Tree Damage Type Classification Based on Forest Health Monitoring Using Mobile-Based Convolutional Neural Network

Abstract

One of the fundamental parts of surveying forest health conditions with Forest Health Monitoring (FHM) is to visually assess the damage experienced by trees under certain conditions. This visual assessment can be facilitated using a Convolutional Neural Network (CNN) which involves building the MobileNetV2 model architecture. The model was trained using 1600 image data with 16 classes. The image data was pre-processed by resizing it to 224x224. The data was categorized into three categories: 80% was allocated for training, 10% for validation, and testing with 10% also. Training was done by changing the values from batches with a maximum of 100 epochs. The model was then incorporated into a mobile application using TensorFlow Lite and testing the application gave satisfactory results. The model results get the best accuracy rate of 98.75% and a loss of 0.0497. This research concludes that the classification of tree damage types based on FHM with CNN can be done. For accurate results, the image provided by the user must be clear and reflect the actual damage observed on the tree.