Automated semantic segmentation and Diagnosis of Coconut Tree Infection Using Drone Imagery and Machine Learning approach for smart agriculture

The coconut industry of our Indian nation contributes an extensive range of economy to the national GDP. About 19,247 million coconuts, or 31.45% of the world’s production, were produced in India in the year 2021-2022 making it the top producer. However, the coconut farmers face major challenges from a plethora of pests and diseases of coconut trees, and yield loss, which affects the farmers economically, standing around at $\mathbf{1 0 - 1 5 \%}$. Major pests, such as Rhinoceros beetles and red palm weevils, are highly destructible, in some scenarios leading even to the mortality of crops. In most of the cases, when trees that are affected need a cure, farmers have to climb the trees to put chemicals in crowns infested with the pests, exposing the farmers to harmful chemicals and risking severe injuries or permanent disabilities. As a solution, the proposed methodology involves the collection of extensive drone imagery from coconut plantations, followed by the annotation and preprocessing of the data. This work utilizes state-of-the-art deep learning models, including Unet, convolutional neural networks (CNNs) for semantic segmentation, to differentiate coconut trees from the background and other vegetation. Subsequently, machine learning classifiers are employed to diagnose diseases by analyzing visual symptoms present in the segmented trees. The proposed approach offers significant improvements in accuracy and efficiency compared with other traditional manual methods, providing real-time insights into tree health and disease management. This system has the potential to enhance crop monitoring, optimize resource allocation, and ultimately increase agricultural productivity.