An Intelligent Deep Learning Framework for Automated Pest Detection in Coconut Trees

Rapid increase of pest associated yield losses in
coconut plantations has led to the acute need for the development
of scalable, accurate and field-ready diagnosis tools. Traditional
manual methods of inspection are labor-intensive, subjective and
are often inadequate for early detection of destructive pests such
as the rhinoceros beetle larva, red palm weevils and eriophyid
mites. To overcome this problem, an intelligent deep learning
framework for automatic pest detection in coconut trees is
proposed in this study, which takes advantage of a hybrid CNN
Transformer architecture, optimized for aerial imaging, taken
using drones and aerial-based imaging devices. The model is
equipped with advanced image augmentation, multi-scale feature
extraction and localization using attention mechanism to make it
more robust in complex plantation environment with occlusion,
variable lightening conditions, and canopy density. Extensive
experiments performed on a newly curated dataset of 14,250
annotated coconut tree pest images show that it can give superior
performance (97.8% detection accuracy, 96.9% precision, 97.3%
recall, and 0.98 mAP) as compared to ten state-of-the-art
baselines. The results validate the potential of the framework to
be used in real-time in precision agriculture systems, allowing
early intervention and less pesticide use as well as better crop
management. Concluding results show that intelligent pest
detection through deep learning can be very beneficial for
sustainable coconut farming and decision support capability for
large-scale plantation monitoring.