Drone-Based Surveillance of Palm Tress Ecosystems

Abstract

This paper presents a novel surveillance system designed to identify the health status of oil palm trees by leveraging MATLAB object detection and deep learning techniques. The study aims to improve the accuracy and efficiency of palm health detection by integrating MATLAB's initial object recognition with advanced deep learning algorithms. The initial phase of the research focuses on elucidating the challenges associated with detecting palm tree health issues using conventional image processing methods in MATLAB. Results indicate that traditional MATLAB object detection methods encounter difficulties in accurately identifying palm tree crowns and assessing their health status due to various factors such as the complexity of crown morphology, lighting variations, environmental conditions, limited feature discrimination, reliance on handcrafted features, and challenges in adaptation and generalization. Subsequently, the study proposes a second stage to enhance the accuracy and efficiency of palm tree health detection through the implementation of a deep learning approach using Faster R-CNN, addressing the limitations identified in the initial phase. Analysis of experimental results demonstrates a rapid increase in accuracy to nearly 100% early in the training process, indicating efficient learning and classification capabilities of the model. Moreover, a significant decrease in Root Mean Square Error (RMSE) at the outset of training signifies a reduction in prediction errors, followed by stabilization at a low level, suggesting that the model's predictions closely align with actual targets in the training data. Furthermore, the loss graph exhibits a similar trend to the RMSE graph, corroborating the effectiveness of RMSE as a common loss function for regression problems. Overall, this research contributes to the advancement of oil palm tree health detection systems, providing valuable insights for future developments in agricultural surveillance and monitoring technologies.