Detection and prediction of wheat pests in complex agricultural environments with an advanced intelligent framework

Wheat is a crucial staple crop globally, and precise field-based pest detection is critical for precision agriculture, sustainable pest control, and environmental conservation. Nonetheless, dependable identification continues to be challenging due to the diminutive sizes of pests, intricate backdrops, and discrepancies among developmental phases. This research introduces an original dataset called the Wheat Greenbugs Leaf Threat Pest form (WGLTPF), which encompasses the entire growing phase and incorporates initial benchmark evaluations. An Automated Image Processing Improvement Form (AIPIF) is further proposed to enhance image quality, suppress noise, and reduce computational cost. Building on these components, a lightweight detection model, P-YOLOv12s, integrated with a CR-ELAN module, is developed to improve feature extraction efficiency while maintaining real-time performance. This phase was followed by developing mobile apps for wheat pest and disease augmented reality. Experimental results show that the proposed method achieves an mAP50 of 92.36% on the greenbug dataset and accuracies of 95.43%, 91.60%, and 88.80% on the standard datasets D1 (97C), D2 (10C), and D4 (1C), respectively. The proposed approach outperforms baseline models in both accuracy and efficiency, demonstrating strong potential for real-world wheat pest monitoring and sustainable agricultural applications.