Switcher-HNet: A switchable hierarchical network for tree species classification from forest stand to individual tree tasks

Switcher-HNet Project & DemoThis project contains code for training, inference, and visualization of tree species classification models using the TreeSatAI and Hierarchical Individual Tree (HIT). Detailed instructions and function annotations are included in the code.The demo is available in Switcher_HNet_Demo.mp4Hierarchical Individual Tree (HIT) DatasetArea: Maoer_mountain_HSI_CHM_Field_dataThe HIT Dataset (Maoer Mountain area) is a multi-source ecological dataset collected in the Maoer Mountain region, a biodiversity-rich area located in Harbin, Northeast of China.This dataset supports research in forest ecology, species distribution modeling, remote sensing, and biodiversity conservation. It integrates high-resolution hyperspectral imagery (HSI), canopy height models (CHMs), and ground-validated field data to enable detailed tree species identification and forest structure analysis.The HSI is characterized by a spatial resolution of approximately 0.2 meters per pixel, georeferenced to the WGS84 geographic coordinate system. It covers a spectral range of 387.8 to 1025.5 nm, distributed across 150 contiguous bands with an average spectral bandwidth of 4.28 nm.The CHMs are a single-band raster dataset aligned with the HSI, encoded in 32-bit floating-point format. It has a ground sampling distance of 0.250 × 0.250 meters and is georeferenced to a projected coordinate system based on the WGS84 datum.The field data include the primary tree species of Northeast China, such as Pinus koraiensis and a variety of broadleaf species.