A hybrid AI-agent architecture for pest detection and digital crop advisory in vegetable cowpea

Vegetable cowpea is widely cultivated across tropical and subtropical regions and serves as a key crop for food and income security. Pest incidence caused by pod borers and pod bugs, remains a major constraint to productivity and the absence of timely detection often leads to considerable yield loss. Current monitoring practices rely largely on manual scouting, which is labour-intensive and may not provide the rapid feedback required for effective intervention. To overcome these limitations, an agent-based pest detection and advisory system, the cowpea pest detection bot, was developed to support real-time surveillance and management. A curated image dataset of major cowpea pests was assembled and processed using Roboflow to standardise inputs for model development. Object detection models were trained using the YOLOv8 architecture and the model demonstrating the best detection performance was selected for deployment. This model was incorporated into an interactive bot accessible through the web and telegram platforms, enabling users to submit field images for automated diagnosis. Following identification, the system retrieves and communicates pest-specific management recommendations derived from the package of practices recommendations of the Kerala Agricultural University, ensuring that advisories remain accurate and locally applicable. The study demonstrates the feasibility of integrating deep learning with agent-based decision systems to create an automated and accessible tool for crop protection. The approach enhances the speed and reliability of pest management decisions and offers a scalable framework for broader digital agriculture applications.