Skip to content
Conference

PSY-Net: an improved lightweight YOLO framework for pest and disease detection in complex agricultural scenes

Oct 2026 · The International Conference on Optical Imaging and Deep Learning · Vol 14494, pp. 144940F - 144940F-12 · 0 citations · 20 references
Engineering

Abstract

Agricultural pests and plant diseases are critical factors threatening global food security. Although deep learning has significantly advanced agricultural image analysis, accurate localization and recognition of multi-scale and densely distributed targets in complex natural backgrounds remain challenging. Furthermore, existing models often struggle to balance detection accuracy, computational efficiency, and class imbalance. To address these issues, this paper proposes an improved lightweight object detection framework named PestSense-YOLO (PSY-Net). Based on a modified YOLO11- Nano backbone, the model integrates a Multi-Stage Training (MST) strategy, a Pest-Aware Augmentation Module (PAAM), and Progressive Image Resizing (PIR) mechanisms. Experimental results on the AgroPest-12 benchmark dataset—characterized by significant long-tailed distributions—demonstrate that PSY-Net achieves outstanding performance with an mAP@50 of 0.953 and an mAP@50-95 of 0.730. The model maintains a superior balance between precision and recall while remaining computationally efficient for edge deployment. This study provides a robust and explainable technical solution for automated, real-time agricultural monitoring systems.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.