Agentic AI-Based Pest Monitoring Framework for Sustainable Agricultural Productivity and Resource Optimization

Abstract

The impact of agricultural pests is significant, on productivity, and the indiscriminate use of pesticides generates production costs, environmental pollution and losses in natural resource use. Current pest monitoring practices mainly focus on classification based on images with no consideration of future pest risk, autonomous decision making or resource optimization. The Agentic AI-Based Pest Monitoring Framework proposed in this study integrates multimodal perception, transformer-based pest recognition and severity assessment, agentic risk prediction, sustainability-aware resource optimization, autonomous intervention planning and closed loop reinforcement learning. The experiments with the IP102 dataset proved the proposed framework to be accurate, precise, recall and f1-score as 98.42%, 98.17%, 97.93% and 98.05% respectively. This framework significantly reduced the use of pesticides by 24.8%, water usage by 17.6% and intervention cost by 21.3% from the conventional management. A Pest-risk prediction score of 96.84% ROC-AUC and a cumulative intervention reward of 0.91 were obtained. These results show the potential of agentic agricultural intelligence to offer the right pest monitoring and increase the efficiency of resources usage, sustainability and effectiveness of autonomous intervention.