A model for forest type identification and forest regeneration monitoring based on deep learning and hyperspectral imagery

Improved Identification of Forest Types in the Loess Plateau Using Multi-Source Remote Sensing Data, Transfer Learning, and Neural Residual Networks

“…The pre-trained and fine-tuned models, as the core parts of the entire framework, are especially beneficial for achieving advanced results in image classification when the target tasks in both phases are the same [16]. In recent years, some studies have successfully identified forest tree species and achieved good classification results by adopting advanced deep learning architectures [20,27,46]; other scholars have also recognized forest tree species by applying transfer learning strategies [47][48][49]. While these specific classification tasks employ models such as fine-tuned transfer learning, they often overlook the quality and relevance of the metadata sets in the pre-training steps.…”

“…In forest monitoring and classification, emerging UAV data (hyperspectral cameras, RGB imagery, oblique photography, and LiDAR) have significantly improved classification accuracy and efficiency due to their high flexibility and resolution [25][26][27]. Many studies focus on using transfer learning strategies, training these models on large and diverse datasets (such as ImageNet) as a pre-training source, and then fine-tuning them on specific small sample datasets.…”