Revolutionizing Pest Detection for Sustainable Agriculture: A Transfer Learning Fusion Network with Attention-Triplet and Multi-Layer Ensemble
Nazmul Haque, Anwar Hossain Efat, S. M. Mahedy Hasan, Nahrin Jannat, Mahjabin Oishe, Mostarina Mitu
Abstract
Pests, insidious and destructive, pose a significant threat to agriculture, stored goods, and the environment. Detecting and mitigating their impact is crucial for safeguarding crops, food safety, and economic stability. Automated detection systems have emerged as critical tools, offering rapid and precise identification for timely interventions. However, existing solutions grapple with limitations, such as extensive labeled datasets and struggles in adapting to new pest species. This study aims to overcome these challenges, developing a robust pest recognition system setting new standards in accuracy and adaptability. Meticulous dataset collection and preprocessing form the foundation for accurate pest recognition. Leveraging pre-trained models and fine-tuning contribute significantly. Integrating Transfer Learning Fusion (TLF) harmoniously blends deep features, and Multi-Layer Ensemble (MLE) techniques enhance performance, achieving 98% accuracy. Our approach resulted in a model with the highest accuracy and precision in pest recognition, surpassing previous limitations. Our work signifies progress in agriculture, emphasizing the importance of automated pest detection and providing a superior solution. Contributions mark a significant stride in safeguarding crops, preserving food safety, and bolstering agricultural sustainability.