A New AM-LBCNN Model for Tree Species Texture Extraction Using Airborne Images

Capturing subtle texture variations in remote sensing (RS) images with limited samples remains a challenge for accurate tree classification. To address this, we propose local binary convolutional neural networks with attention mechanisms (AM-LBCNNs), a deep learning framework that convolutional block attention module (CBAM) with local binary convolutional neural networks (LBCNNs). Experimental results on a limited dataset of remotely sensed tree images demonstrate the superiority of AM-LBCNN in capturing subtle texture differences, achieving a classification accuracy of 93.57%. The processing efficiency of LBCNN has been improved by five times compared to the baseline, and after introducing the CBAM mechanism, the Top-1 accuracy has been increased by 0.11% compared to LBCNN. This integrated approach addresses the limitations of CNNs in texture extraction by using LBCNN and the attention mechanism (AM) to accurately identify fine-grained texture variations among tree species, while mitigating the computational challenges associated with few-shot.