A Unified Deep Learning Framework for Accurate Pest Detection and Classification in Agriculture

Abstract

Introduction: Agriculture is an important role in sustaining human life and ensuring high quality food production is essential for economic growth. Among the main difficulties farmers encounter the rapid spread of insect and pest infestations which can significantly impact crop yields.

Objectives: While, existing approaches have explored pest detection and classification, often suffer from inaccuracies and inefficiencies. To address these issues, this paper propose a unified Approach for PEST detection and classification model called SAMYNET (Segment Anything Model + YOLO8 + EfficientNet system).

Methods: SAMYNET an advanced automated system in that combines three cutting-edge DL models: Yolov8 for object detection, Segment Anything Model (SAM) for image segmentation, and EfficientNet for fine-grained classification. The system runs in a pipeline with several stages..

Results: To begin with, YOLOv8 is used to identify probable pests in the crop photos and create bounding boxes around them. SAM then processes these identified areas of interest (ROIs) to generate accurate segmentation masks for every pest, defining their precise borders.

Conclusions: Finally, EfficientNet classifies each segmented pest into specific categories providing detailed identification. Experimental results demonstrate that SAMYNET outperforms traditional methods in pest detection, segmentation and classification achieving high accuracy of 95% and precision 89%. This automated system offers farmers and agricultural experts a scalable, efficient tool for timely pest management which ultimately crop yields and promoting sustainable agricultural practices.