DEEP Q-GUIDED GENETIC EVOLUTION WITH HYBRID LSTM-CNN FOR TREE SPECIES IDENTIFICATION IN REMOTE SENSING

Accurate identification of tree species from satellite imagery plays a crucial role in sustainable forest management, biodiversity monitoring, and ecological assessment. However, the heterogeneous nature of satellite data, high spectral variability, and noise interference present significant challenges for robust classification. To address these limitations, this study proposes an advanced hybrid framework for tree species identification that integrates preprocessing, multi-level feature extraction, and intelligent feature selection with deep learning-based classification. Initially, raw satellite images are enhanced using histogram equalization, non-local means denoising, and data augmentation to improve quality and reduce noise. For feature extraction, both handcrafted and deep learning-based techniques are employed, including Local Binary Patterns (LBP), Gabor filters, ResNet-50, and spatio-spectral CNNs. Dimensionality reduction is further refined through Kernel Entropy Component Analysis (KECA) for hyperspectral data and Spectral–spatial Coherent Discriminant Analysis (SCDA) to preserve discriminative spectral-spatial information. To ensure optimal feature selection, a novel Deep Q-guided Genetic Evolution (DQ-GE) strategy combines the exploration ability of genetic algorithms with the adaptive learning of deep reinforcement learning. The selected features are then classified using a hybrid LSTM-CNN model that captures both spatial patterns and sequential dependencies, while an ensemble model enhances robustness and generalization. Experimental evaluation demonstrates that the proposed architecture significantly improves classification accuracy, species differentiation, and computational efficiency compared to conventional models. This framework offers a scalable and reliable solution for automated tree species identification in large-scale remote sensing applications.