Optimized Multi-Component Attention Graph Convolutional Neural Network for Detecting Invasive Mangrove Plant Derris Trifoliate from UAV Imagery

The invasive mangrove species Derris trifoliate poses a serious threat to coastal ecosystems by disrupting native biodiversity, altering soil properties, and impacting hydrological patterns. Accurate detection of this species is challenging due to its spectral similarity with native mangroves and the complexity of remote sensing data. To address these challenges, an Optimized Multi-Component Attention Graph Convolutional Neural Network (MCAGCNN-DIMPDT) is proposed for the precise identification and monitoring of Derris trifoliate. Initially, UAV (Unmanned Aerial Vehicle) imagery is collected using an RGB (Red, Green, Blue) camera to capture images of Derris trifoliate in mangrove forests. In the pre-processing stage, images are prepared and resized using the Multiple Local Particle Filter (MLPF). The pre-processed images are then processed using the Multi-Objective Matched Synchro Squeezing Chirplet Transform (MOMSSCT) to extract texture features. These features are fed into the Multi-Component Attention Graph Convolutional Neural Network (MCAGCNN) to detect Derris trifoliate and classify areas as Derris trifoliate or other plants. To enhance classification accuracy, the Adaptive Lightning Attachment Procedure Optimizer (ALAPO) is employed to optimize the network’s weight parameters, enabling precise detection of Derris trifoliate in complex mangrove environments. Experimental results demonstrate that the MCAGCNN-DIMPDT framework achieves an accuracy of 99.47% and a precision of 99.35%, outperforming existing detection methods. These results demonstrate how well the proposed method detects and categorises Derris trifoliate, facilitating better invasive mangrove species monitoring and management in coastal ecosystems.