Leaf area index (LAI) is one of the key biophysical parameters in forest ecosystem modelling, e.g. in yield and carbon sequestration assessments. LAI is costly to assess in situ, while full-waveform LiDAR (wLiDAR), which captures detailed vertical canopy structure, has proven effective for LAI estimation in forested ecosystems. However, recorded waveforms are inherently broadened by the system impulse response, reducing temporal resolution and structural fidelity and consequently biasing derived canopy metrics. Building upon our previously-established empirical LAI modelling framework, this letter investigates the impact of waveform deconvolution as a preprocessing step to mitigate system-induced distortion and estimate the underlying target cross-section, prior to metric extraction for LAI modelling. The Gold deconvolution algorithm was applied to airborne wLiDAR and evaluated in both simulated and operational datasets. Deconvolution enhanced peak-to-valley contrast, improved canopy-ground separation, and strengthened metric-LAI relationships. For the real LVIS (Land, Vegetation, and Ice Sensor) data, deconvolution increased the full-sample adjusted R2 ${R<^>2}$ from 0.56 to 0.64 and reduced the leave-one-out cross-validated MAE from 0.35 to 0.29 when integrated into the empirical LAI model. These results indicate that waveform deconvolution is a physically interpretable preprocessing step that may modestly enhance the performance of empirical LAI models for operational airborne LiDAR data.
Although anchor-free detectors have shown promise, they still struggle with insufficient multi-scale feature representation and limited geometric adaptability when detecting tiny objects with drastic scale variations and complex geometric structures in remote sensing images. To address these challenges, we propose a fusion-perception framework consisting of a Progressive Fusion-Perception Module (PFPM) and a Dynamic Deformable Perception Head (DDP-Head). PFPM performs progressive multi-scale feature aggregation across channel-split sub-feature groups, while DDP-Head integrates spatial attention, deformable convolution, and dynamic convolution for adaptive prediction. Experimental results demonstrate that the proposed method achieves a 2.7% AP improvement over the baseline on AI-TODv2 and also improves detection performance on LEVIR-Ship.
Precise classification of tree species has significant value for urban forest structure and function research. However, current urban tree species classification is limited by complex urban backgrounds and the suboptimal performance of single classifiers. This study selected a forest park in Zhengzhou City, Henan Province, China, as the research area. The study integrated 10 m spatial resolution hyperspectral satellite imagery, 1 m spatial resolution GF-2 panchromatic satellite imagery and Unmanned Aerial Vehicle–Light Detection and Ranging (UAV LiDAR) point cloud data. Different feature combination were constructed by extracting features, and a multi-classifier ensemble algorithm was employed to classify tree species. The results showed that: (1) Compared to single-source data, multi-source data fusion achieves higher accuracy. Especially, spectral and texture features combined with canopy height features yielded the highest classification accuracy. The overall accuracy (OA) was 94.59% and the Kappa coefficient was 0.9336. (3) Compared to a single machine learning algorithm, the weighted voting ensemble learning algorithm had superior classification result in the study area. The OA improved by 3.22% and the Kappa coefficient increased by 0.0441. This study demonstrates that multi-source remote sensing data can achieve feasible accuracy for tree species classification in complex urban surface environments.
