Ensemble learning based on neural networks for tree image segmentation

Tree image segmentation is an essential component of forest management. Many convolutional neural networks (CNNs) based methods have been proposed for the image segmentation task, but people need to adjust those network architectures and parameters to achieve good performance when addressing a new task, which usually requires expertise in the problem domain and extensive experimentation. Ensemble learning is a technique that combines multiple models to make more accurate and robust predictions, where the key is the accuracy and diversity of the base models used to construct the ensemble. Therefore, this work investigates an ensemble learning method based on CNNs for real-world tree image segmentation. Our results demonstrate that the ensemble model outperforms individual models regarding both the qualitative and quantitative results in segmenting tree crowns from the background in the images taken from Wellington, Aotearoa New Zealand.